Good, coding harnesses should be open source and LLMs should be treated as commodities. Minimize switching costs for consumers, and let people understand how they're interacting with the context and the LLM outputs.
The industry has been moving the wrong direction with Claude Code staying closed (despite multiple times leaking the source code!) and the open source Gemini CLI being deprecated in favor of closed source Antigravity CLI.
Why would a company do any of these things? What is their motivation for any of it? That’s like saying cloud providers should be commodity and should open source all of their platforms and eliminate egress fees so customers can easily leave at any point in time.
On the question of language models and periphery tooling -
Open weight models are disruptive to the business models of closed model businesses. An incentive is if your business is built around X but model training is helpful to you, but you don’t expect to meter it specifically. You can release your models and undercut the exclusive moat of a new model company like OpenAI or Anthropic from becoming at some point a competitor, or holding their access as a chip in pricing negotiations. By opening your architectures and weights other competitors can build on them and newer better models emerge faster decoupled from a small number of proprietary models. This lets you focus on X while gaining overall momentum on your model release at no additional cost and no loss in focus on X, while defending against upstarts and monopolies.
This is effectively a lot of the open source world that comes from corporate development as well. It feels odd after this many decades of discussing corporate reasons to participate in open source we keep rehashing it.
Having a five-year plan dictating which publicly owned b2b SaaS AI service you will be provisioned sure sounds like a dream of mine. I wonder what could go wrong.
Open Source comes out of Anarchist philosophy not Communist State owned type ideas. There is no direct ownership, its a project built as a common good, and owned by "the public". Its a community garden that people are free to plant and harvest as they see fit.
Yes I do think cloud providers should open source all of their platforms, and this is not charity because it is essentially the hosting that they are providing as a product. Even if, say, google open sources its whole search infrastructure, it does not at all means you can just host your own due to the huge hardware requirements, but you can know(especially after AI which can be utilized to do this) that they are not using your data in a way they shouldnt.
> because it is essentially the hosting that they are providing as a product
That would be true _if_ they were forced to open source all their code, but it isn’t today.
> Even if, say, google open sources its whole search infrastructure, it does not at all means you can just host your own due to the huge hardware requirements
You can’t but Acme Inc and other bigcorps could, and Google’s margins would evaporate overnight.
It's because they do things that is why they score differently. Coding hardness add features for user experience not for agent efficiency. If they did all the coding hardnesses would be using bash and code mode and letting the agents write code to perform tasks but this doesn't work because you want humans in the loop. You want users to be able to approve and deny writes. You want uses to see edits. So you have to build tool for these. It's hard to show diffs when the agent is just using bash.
> The added features are just user experience features.
> It's because they do things that is why they score differently.
That was my point. Regardless of how you feel about UX, it's a value added set of features. The question initially posited, stands. Why would a company do any of these things?
> Coding hardness add features for user experience not for agent efficiency.
Pretending it was always about some metric you just decided was important is moving the goalpost. It's not compelling.
I think it makes more sense that it's Freemium Dominance or they act as Low-Cost Marketing tools.
Try Kimi in Kimi CLI and Claude Code and try saying that again. Kimi quickly collapses into tool calling loops without measures in their CLI but not in Claude Code and is largely useless for any long running tasks in harnesses not taking this into account.
With those measures (which are actually quite interesting) it can at times perform at Sonnet level.
The reason I have a dog harness is to distributes weight so I don't choke her when she goes at the other dog that she doesn't like. I'm actually puzzling over kids kamikazeing into cars
It's actually only a problem if it's the other way around, isn't it?
If kids run into a car, they will most probably just bounce and continue, perhaps inflicting some minor damage. But if a car mows down a kid, that could well be a fatal injury. Leashes for all the cars! ;)
Building a good and working coding harness with smaller models is really hard. Everything evolves around the limited context size.
Tools must be specification driven to reduce noise and high temp hallucinations, tool call shrinking needs to remove errors and tryouts of different formats of parameters (because LLMs always ignore descriptions in the JSON...), and you have to deal with long running agents because you can't afford them. Planner/orchestrator architecture, agent to agent communication need to be summarized, and then you have the messed up scheduling parts, because you need to prioritize short running agents and give the planner a tool to wait for outputs of spawned contractor agents.
And that's not even talking about sandbox vs playground read/write/access policies of tools.
Harness engineering, if done correctly, is quite hard.
And all of this works 60% of the time, every time.
Anyways, that was somewhat the summary of the last 6 months building my exocomp agentic environment. And it's still not satisfying to work with.
In my limited experience, the smaller the model, the bigger the harness. Where with something like claude or deepseek the context size etc just let's you give it bash access and step back; small models tends to do better with simple action - response , new context each call. Context management becomes a continuous activity. Its a fun space , and I have found big models decent at building and improving these harnesses for the small ones. Using /loop and just run a continuous test - build - test loop.
Cloud providers are commodity, and egress pricing is partially cost following because they have to pay peering to their interconnect points for WAN. Internal networks are not charged within the account because the economics of the VPC overlay are optimized for that use, but inter account and VPC and other boundaries carry cost - especially interconnection between accounts because the way VPC treats virtualization requires a relatively expensive routing. Inter AZ and inter region pricing also exists for the same cost following reasons. They also help shape incentives because it allows them to optimize placement of compute within the same AZ to physical buildings or rings.
The case that is largely nonsense is the egress pricing on direct connects since beyond the circuit costs, which the customer pay, there’s no costs for aws not already on the customer regardless. It also makes DC friction weird in that you are incentivized to NOT move storage before compute.
opensourcing software may enable leverage of wider network of contributors to given piece of software,hence software can evolve much more quickly and efficiently.
Joel Spolsky in 2002 identified a major pattern in technology business & economics: The pattern of "commoditizing your complement", an alternative to vertical integration, where companies seek to secure a choke point or quasi-monopoly in products composed of many necessary & sufficient layers by dominating one layer while fostering so much competition in another layer above or below its layer that no competing monopolist can emerge, prices are driven down to marginal costs elsewhere in the stack, total price drops & increases demand, and the majority of the consumer surplus of the final product can be diverted to the quasi-monopolist.
No matter how valuable the original may be and how much one could charge for it, it can be more valuable to make it free if it increases profits elsewhere.
This pattern explains many otherwise odd or apparently self-sabotaging ventures by large tech companies into apparently irrelevant fields, such as the high rate of releasing open-source contributions by many Internet companies or the intrusion of advertising companies into smartphone manufacturing & web browser development & statistical software & fiber-optic networks & municipal WiFi & radio spectrum auctions & DNS: they are pre-emptive attempts to commodify another company elsewhere in the stack, or defenses against it being done to them.
"Commoditizing your complement" is a convenient market phenomenon when it happens, but it's not a hard rule that everything that should be a commodity lands in the window of incentives that result in this strategy being the self-serving optimum for a company. Usage of "should" here can be interpreted as "would be most beneficial to the largest number of people in terms of cumulative productivity" or "not having this be commoditized leads to terrible incentives that harm a lot of people." There's no guarantee at all that it ends up being most profitable to individual market actors to attempt to commodify LLM compute, and in fact they seem more interested in securing legal moats and protected monopolies. You may think that reality will force this to fail via open source model releases from China or the like, but consider the world in which the US officially bans import and usage of all but specifically licensed and approved LLMs. The most powerful companies and people here seem broadly aligned on wanting that, so it's not implausible. Sure, you as an individual could illegally import and run an unsanctioned LLM yourself, but most people won't do that, and attempts to scale it would be beaten down with the force of law. The companies who strike deals will get 99+% of business and profit handsomely, no need to ever let their golden goose be commodified. It's a strictly better strategy if your government is dysfunctional enough to let it happen.
> but consider the world in which the US officially bans import and usage of all but specifically licensed and approved LLMs.
Then the US will fall behind China and, more importantly, from the rest of the countries in the world who refuse to lobotomize themselves in the manner you described.
There are also other cultural and political risks which, although delayed, may easily grow into something apocalyptic. Rich people building bunkers isn't just a fad, the outcome they envision isn't fiction, it might actually be a part of the plan, and while this is concerning it does make everything simpler.
I don't really see how this applies to Anthropic. Claude Code _is_ their integral money maker. Their point above is a three-way neck-to-neck competition where they won't be able to become a near-monopoly any time soon.
Anthropic was making moderate revenue from their API tokens, but everything changed when they started bundling crazy amounts of usage credits at $20/$100/$200 claude max subscriptions relative to the API token rate, but only as long as you use their coding harness — despite claude code, back when this started at least, being an inferior coding harness compared to many open source ones.
CC is technically a free product (you can use it with any model). It's got very few popular "opinions" on how a coding harness should work, but it's by far the most popular one (including at competing LLM manufactories like OpenAI and Meta and Google). Why? If it was just that their models are 5% better, most workplaces would optimize on token (aka cost) efficiency.
Anthropic has been winning the usage of their harness, their tokens, while earning significant revenue, by significantly subsiding their token consumption.
This has earned them many things:
- prime data on how software development — simultaneously the leading beneficiary industry from LLM use, and also the most flush with cash to spend — has been using LLMs
- bringing that industry to standardize around their harness concepts. They are essentially establishing themselves as the W3C of LLM interfacing, except as a private organization.
Most of the cloud platforms are open source. Linux, container, k8s… it’s entirely possible for someone to build and deploy their private cloud if they have the resources.
> and eliminate egress fees
What does it mean? If I sign up for cloud service I am only bound to the contract terms. If I am PAYGO I can switch anytime.
> Most of the cloud platforms are open source. Linux, container, k8s…
Linux isn't a cloud platform and neither is Docker.
Kubernetes was created specifically to create a way in against AWS' de facto public cloud monopoly. The Cloud Native Compute Foundation is a classic "alliance of smaller players uses open-source and interoperability as a wedge against an incumbent that threatens hegemony".
> it’s entirely possible for someone to build and deploy their private cloud if they have the resources.
What is there for that, really? Basically just OpenStack?
I never say these pieces are the equivalent of a cloud platform. What I said is that with them, it is entirely possible for someone to build their own.
Why someone wants to do that is not my concern. But not being able to is.
The cloud provider isn't the harness, Terraform/OpenTofu/Pelumi and the abstractions you build using them are. The cloud provider is the LLM. It's not as fungible as the LLM and there's no direct comparison to egress costs of course, but that's moreso a problem with the metaphor.
Because Claude Code is literally nothing particularly special. We don't need their business model. They need their business model, and to that I say, tough shit.
China is giving away AI for free so it’s harder to make money. The same strategy they did with solar panels. Sell them at a loss long enough until the manufacturers go out of business and you’re the only one surviving. Then flip to extract monopoly profits
Instead we got home solar system become very affordable over the past decade.
And driving out US manufacturers isn’t even the main goal for China. They know their huge risk on reliance on petroleum and was doing everything they can to mitigate that. Building out a huge solar manufacturing base is their answer. Now they are reducing petro imports YoY.
> China is giving away AI for free so it’s harder to make money.
That might've crossed their minds but that wouldn't move their hand, not even a finger. Politics is the primary driver here, here's the deal:
AI is the new Internet
China foresaw a world where they'd be blocked from it and Anthropic's ongoing attempt to block their own country form it shows how right the Chinese were.
Yes. Releasing open weights seems to be an effort to slow down competition. Facebook was behind. Grok was behind.
Don’t get me wrong, I am glad they are doing it, I personally use open source models. In part to not spend money on the other APIs. So it’s clear that some percentage of what would be paying uses choose to not pay. The other part of opensource is free community labor. How much development work did Facebook get for free around the core infrastructure of react by open sourcing it? A massive amount that they did pay for but benefit from.
The point is, it may benefit society, and yet that social benefit wasn’t the motive for them releasing it as open source.
There is a difference between a loss leader that drives a monopoly. Microsoft gave away a browser to make their monopoly OS more valuable and deepen their anti competitive moat. They were almost broken up as a result and honestly should have been if we had any anti trust enforcement.
How does open sourcing a Claude code clone drive adoption of anything that is a monopoly or even commercially related? Instead it seems like an attempt to undermine US AI companies.
That being said I am increasingly skeptical of how the US leaders are converging on creating monopolies and going deeper into the app layer that means they will end up owning everything rather than being the substrate of a competitive and flourishing ecosystem.
If it were up to me I would implement a regulation that 1) AI labs can’t own inference hardware, data centers would be a regulated utility like electricity and the internet required to provide open access third party safety, guardrails and audit, 2) inference providers can’t build apps beyond serving API requests 3) training data sets are required to be open sourced within 3 years of training a model.
What we are doing now is allowing vertical and horizontal integration of the hardware, training, inference and application layer. Last time we did that standard oil ended up owning the rail network, pipelines, oil fields, gas stations and refineries. Go see how that worked out for society.
Good will and trust can ultimately have monetary value, and having a funnel based on open source is a viable play if it leads to a service that is sticky.
I don’t like changing tools. What engineer does? I want to learn one tool and tune it to my exact preferences. Proprietary vendor tools are not portable and I avoid them.
Either Anthropic or OpenAI could drop the first-to-market open coding harness tomorrow and it would be as big as VSCode, it would be the standard platform everyone builds stuff on.
The familiar Chinese recipe for success: Always copy and imitate first, even if it is inferior, always make it cheaper or even free so that the original innovator will be burdened by brutal price competition and much bigger R&D costs and cannot keep up in the long run. Then the copycat will win in the endgame.
There are very rarely original ideas or original products. It’s more about money put into that idea and the marketing. You could say that Apple copies and imitates just like the chinese. After all they wait for the others to pour money into the ideas and they just deliver a polished experience on top of the hard work others did. Yet people don’t accuse them of being imitators.I think people are mad at the Chinese because they do it for cheap and at an industrial scale on so many products/industries(i.e even food). To me they are a force for good competition.
show me from which country or company did the Chinese copied their EV techs, batteries, drones, robotics etc. for their military gears, show me show they copied their J-36 and J-50 fighter jets, Type-055 destroyers, YJ-21 missile and the most recent 09x sub.
time to stop living your socially isolated life and start reading stuff other than whatever fed to you by your brainwashing media.
After 16 years on Hacker News, I've come to associate its readership with cheap bastards who think everything should be free while simultaneously wanting to keep their 6-figure jobs.
There's a very strong overlap with male gamers, who also think everything involving sophisticated engineering and design should be cheaper than a cup of coffee.
Just call it out and maybe we can collectively choose to towards a culture that doesn't encourage such shameless behavior or perverted values.
> on Hacker News, I've come to associate its readership with cheap bastards
They might be cheap but they surely aren't stupid. They sure know who not to trust, security beats imaginary profits promised by cheap used-car salesmen who did move software jobs to wherever they could spare a dime.
And you want us to defend their current profits which they swear to use for the removal of the remaining software jobs...
If you see this as people just want free stuffs you are missing the point. There is an entire different level of danger to allow a few entities to completely own a day to day technology.
Imagine personal computers and open source operating systems completely don’t exist. What you can do on a pc today, you have to always do it on a thin client always connected to a main frame run by a selected few. Everything you do is recorded and subjected to at least 30 days of retention and inspection.
Imagine every car is completely not customer serviceable and have to be connected to one of the three manufacturers in order to operate. Everywhere you go is recorded. The manufacturer may decided the place you want to go is inappropriate for you to go.
You’re not totally wrong but I think the motivations are varied. Using myself as an example, I have paid huge sums for software, films, news, and yes open source software, but I hate paying 5 bucks for a sub or buying certain products because the deal just isn’t a fair trade. I’d say most modern software is a bad deal.
The jagged frontier of frontier models means treating tokens as fungible between providers is naive at the limit of capability, but also will work for solved problems far from the boundary. The problem is you need to keep evaluating all models to know where your use case lies on the frontier map.
As a concrete example, you’ll get very different results for the same prompt for sonnet, opus, fable, gemini, gpt 5.5, …
There is really nothing free in terms of money, there are only things really free in term of spirit. But AI coding assistant are not those things related to spiritual freedom.
The complements ARE the LLM AND the harness. The actual products we're all consuming are GPUs. Memory being expensive is a second-order effect.
The platform is the GPU, and doing cool shit with it IS the complement, which requires more memory. And demand is so high and will stay high, that it looks like the platform itself.
The question is why supply is restricted, primarily by sanctions and tariffs to China, and the expressed refusal of RAM makers to even think about increasing supply, they are actually all sweaty about China taking a bit of the unrestricted market.
> MiMoCode is built as a fork of OpenCode. It keeps all core OpenCode capabilities (multiple providers, TUI, LSP, MCP, plugins) and adds persistent memory, intelligent context management, subagent orchestration, goal-driven autonomous loops, compose workflows, and self-improvement via dream/distill.
Sounds like they slapped in a bunch of common plugins and released it as a product to promote the free-for-a-limited-time use of their new coding AI service.
> promote the free-for-a-limited-time use of their new coding AI service
Not sure which "free" service you're referring to, but MiMo v2.5 Pro is plenty capable & (after its recent 70%+ price drop) one of the most affordable options in its class (DeepSeek v4 Pro, MiniMax M3, & Qwen 3.7 Plus). I read somewhere that Labs are incentivized to implement custom harnesses because each model has its strengths, quirks, & blindspots (like Qwen forking Gemini CLI)?
I'm guessing the greatest reason behind each provider creating and agent harness is that (a) there is not a clear winner still and (b) it is harder to switch models with a competitor, as you also have to switch harness
Or by using a proxy, yeah. Personally I would still prefer a multi provider harness over CC when using it with another provider, if alone for the visible reasoning, model switcher, cost estimation and so on. So far I've only preferred CC when I needed to work with Jupyter Notebooks because it has built-in tools for that.
Since the link is in Chinese: MiMo Code is Xiaomi’s AI agentic coding harness.
“ MiMoCode is a terminal-native AI coding assistant. It can read and write code, run commands, manage Git, and use a persistent memory system to keep a deep understanding of your project across sessions while continuously improving itself.”
Thanks, I missed that on first glance and did manual translation.
Not sure why my iPhone shows an option to translate website but all the destination languages to pick from (I have multiple languages installed), including English, are greyed out. iPhone does support translating from Chinese (Simplified or Traditional), and the button to translate website isn’t greyed out like it is for unsupported/unrecognized languages. Might be an iOS 27 bug, because it is working on other websites?
It's entirely possible, and even standard, to allow the browser to tell your site which language to respond in.
While ignorance of internationalization standards is a possibility, and the most likely cause.. I do wonder if it's a bit of a nudge to promote Chinese influence in the AI space.
Not that they really need to do that, China is already doing great (relatively, depending on criteria). The implosion of the US, the resulting brain drain and world shake-up has been very timely for their AI and other industries.
It's a very smart move for them to think longer term and start freezing out NVIDIA. Then they can take Taiwan purely for ideological concerns and not worry at all about the fabs blowing up in the process.
And they won't be dependent on foreign factories sitting on an island just off the shore of a superpower who's shown nothing below absolute resolve for decades towards the idea of conquering that island....
You're right, there are probably lots of sites misconfigured to not respect language headers, but we don't notice because English is the default.
However, the right solution is still to use the language header. I send that to them, they should use it to give me the right one by default.
One of the funny things is that this whole site is in an iframe; which breaks both Google Translate, and the Firefox translate feature. If you check, the outer iframe seems to indicate `lang="en'` and loads the iframe with `src="/coder/index.html?lang=en"`, but the inner iframe still gets a `lang="zh-CN'` by default until you use the toggle.
If you go to the eventual redirect source of the page with `lang=en` parameter, you get a `lang="en"` attribute, but it's still in Chinese until you toggle it with the menu: https://mimo.xiaomi.com/coder?lang=en
Anyhow, yeah, lots of pages are probably broken this way but we don't notice. But still, it has that info from your request, it should use it.
But what if you have English configured as a preferred language? Isn't that what it's for? Wouldn't it make sense for a website to respect that (when available)? I hate that google.com doesn't and defaults to random languages based on IP.
Web standard often give great grounds to leverage on. Modern stacks often really poorly work with a lot its surface and reinvente half baked bespoke alternatives.
Personally, none, I’m not English native. I didn’t notice the locale switch, but mostly because the look of the website was so beautiful I didn’t pay attention to menus. I wonder if ideograms keeps looking so beautiful once you learned to decode them. I never found Latin script to be particularly beautiful, and to this date Arabic script remains my favorite one in term of esthetic (I can’t read Arabic ever).
This is not exactly what you’re asking about, but I started learning Japanese when I was in the middle of playing Cyberpunk 2077 (for unrelated reasons); and I gotta tell you; realizing that 98% of the Japanese text everywhere in the game was just “hotel” or “karaoke” definitely took away some charm from it.
When you look at the source, you can set it to English from the params https://mimo.xiaomi.com/coder?lang=en
but there's a small bug, the hero subtitle isn't translated but everything else is.
What a transformation by Xiaomi to build almost frontier level models. Five years back, when I was in the data science team, they dint really bother about AI models and were using Baidu for NLP and vision under the hood of their APIs
I fully expect Baidu and other tech giants on the Chinese shores to try and push the boundaries of technology. Silicon Valley (and the US) in general has always been the hot-bed of innovation. But with enormous increase in wealth in China (and to an extent in India), I can see these companies being more and more ambitious. Not long ago Andrew Ng of Coursera and Stanford AI Lab fame joined Baidu to further their rival to the 'Google Brain' project.
Xiaomi has long been positioning itself as a company with design chops of Apple, engineering chops of Google, and e-commerce chops of Amazon, all rolled into one-- and I can see where they are coming from. If they manage to pull it off, I guess that's when we'd start seeing the proverbial "Death of Silicon Valley" as in, it loosing its strange monopoly and strangle hold on tech world in terms of both talent and innovation.
"Death of Silicon Valley" in this case is such a funny perspective. Like, how twisted is the US's view of the market that they think "Competition? Oh no. Sound the alarms."
Except it’s not competition if US companies can’t access the Chinese market but Chinese companies can access the US market. Just like cars. America is not willing to compete with BYD. But we are 20 years into massive IP theft from China and the naive and short sided leadership in the US that basically traded our knowledge, design and manufacturing knowledge for cheap of shoring, and watch China execute spectacularly to take advantage of the opportunity.
Xiaomi have been cooking a lot in recent times. Their model, especially the pro series, is underrated in my opinion. It haven't received the attention it deserves while it is pushing higher and higher in benchmark scores (looking at artifical analysis), and this was before Deepseek dropped V4.
Furthermore, their pricing plan is insanely cheap, they even upped usage limit for their cheapest plan, lite plan, which is at 5$ / month. And now, they are dropping a Harness for their own model? Amazing. I wish they added support for installation through Homebrew though.
On another note, this is what I would like to see more of from a company, what I do not welcome is startups making their model exclusive and hurt their customer base through sabotaging as a way to prevent eventual distillation attempts.
>Furthermore, their pricing plan is insanely cheap, they even upped usage limit for their cheapest plan, lite plan, which is at 5$ / month.
Unless something changed their plans aren't really worth getting. They're not that much cheaper than the per-token rates, and because it's a plan, you have to contend with weird usage restrictions. You're better off paying per-token unless you have some use case that demands a very steady stream of tokens.
Looks like they have very effective collaboration with DeepSeek and Kimi. Those three models have been bouncing ideas and sharing R&D innovation, which made all of them improve very fast.
Based solely on quality and price, OpenAI, Anthropic, and other western models just can't compete with the new generation of Chinese open models.
>Looks like they have very effective collaboration with DeepSeek and Kimi.
The collaboration is informal. People don’t seem to realize this, but the Chinese internet for programmers and developers today feels a lot like StackExchange in its heyday. There’s a huge emphasis on sharing knowledge, because sharing what you know builds your profile, and becoming a rockstar in a subfield is one of the only ways to get ahead.
Competition in China is ruthless. But unlike in North America, where individuals are often bound by agreement to hoard knowledge because it can give them a competitive edge, the competitive advantage in China is building face and peer recognition. And that comes from proving that you are worthy of being a "master/teacher", and that extends to the valuation of your knowledge business. For example, the third wave coffee shops in China, the master roaster is often called "master/teacher" once they win a roasting competition and start sharing new knowledge of roasting in the public sphere, and that's a title of sincere respect.
You can see parallels with those that apply to give talks at conferences and post snazzy technical presentations they give in the US, but the bar for what qualifies as new knowledge is far higher in China because there's a massive ecosystem of people rushing to outcompete what you have to offer, and once the ball gets rolling on knowledge sharing, lots of people will go off and build upon that knowledge or try to build businesses on top of that, which in turn produces more knowledge.
Reading developer forums in China, once you crack the code (I find Gemini will get you a good chunk of the way with good translations), they are really quite far ahead with what they're willing to share. And I suspect in great part, the decision to release open-weights is heavily tied to that concept of building face/peer recognition = building valuations.
Care to share a few such forums or websites, genuinely curious as someone who reads basic Chinese.
A decade ago, I knew there were CSDN, jvjin (may not be the correct pinyin) where people talked about tech. Nowadays they have just become SEO farms like the rest of the internet.
Actually there are very few Chinese forums left, because it is a huge hustle to run one, and the same rule applies to the comment section of any personal blogs.
Right now pretty much the only one left with a bit discussion is v2ex, but it's blocked by gfw and kinda becoming a ranting corner for programmers. Some also recommend nodeseek and a couple other new sites, where you can find shady stuff but not tech stuff. Then you have tech blogs of big corps, or personal ones, but still not many quality tech posts. There are also some on WeChat but it's kinda cumbersome to use so I stopped.
Thank you for the insights, this sounds like an environment I want to partake.
> Reading developer forums in China, once you crack the code (I find Gemini will get you a good chunk of the way with good translations), they are really quite far ahead with what they're willing to share.
Are you able to share those forums and other resources? I would love to read what people in these communities are sharing.
> Looks like they have very effective collaboration with DeepSeek and Kimi. Those three models have been bouncing ideas and sharing R&D innovation, which made all of them improve very fast.
Very fascinating to learn this, didn't know Moonshoot (Kimi) also collaborated with others. I think I read in another post that DeepSeek and Qwen team shared the same building? So that kind of explains it.
> Based solely on quality and price, OpenAI, Anthropic, and other western models just can't compete with the new generation of Chinese open models.
I have to agree. I had the great opportunity to take the offer Z.ai had with their Christmas deal, their lite plan was 3 months for 7$. GLM-4.7 was already impressive enough.
When they released GLM-5-Turbo and GLM-5.1, that is when I came to the realization of how close the gap is between proprietary western models and Chinese open-weight ones (not all of them are ofc).
I could barely believe how good GLM-5.1 was, I didn't think I was using it in CC and had to check the settings again. It's astonishing how close the gap is now, and this competition benefits us very much, the pricing is so low atm, its amazing.
Pretty neat that you can just install it and start using it (at a Sonnet 4.6-level model) without needing to sign in or pay.
Typically, Chinese websites are a big pain to log in or sign up because they require a +86 phone number due to legal reasons. Being able to use it without having to make an account is amazing for friction reduction. I could probably even just install it onto new machines to help with set up.
I wonder how they are gonna detect and block abuse though?
MiMo v2.5.0-Pro is honestly the first Chinese model that I've tried where I really though why should I use Claude Sonnet when I can get the same results for a fraction of the cost. There was always something off about Chinese models that made it apparent that it couldn't fully compete with GPT, Claude, Gemini, etc. but this was the first model where I was like, this feels like Sonnet.
I can't prove it, but I think they trained heavily on Claude output. From my perspective I don't care since Anthropic trained on my data.
Using them also works well for North Americans as our peak hours is not theirs.
If I had one complaint, the v2.5.0-Pro model thinks too much.
They won't be giving this away, at least not for some years. It almost certainly has distillation data embedded in it, and that would be a smoking gun.
GLM 5.1 is stronger than Sonnet 4.6 in my opinion, but while they have a coding plan that is a good value MiMo beats it on price. I haven't used MiMo much yet but it felt pretty similar.
So funny I have noticed how terrible the signup is on all these Chinese models, companies etc. Always wonder why it is such an easy process. Like QQ, Tencent etc demos Ive seen past year
Claude and Codex pricing will eventually have to come down, for most common coding tasks you don't need a super smart slow model but a smart-enough and very fast one.
Microsoft github copilot recently changed their billing. i'm on the yearly subscription. GPT-5.4 is now 6x and even previously free model like GPT-5 mini now cost .33x. its only June 11 and my usage is now at 50%.
I don't think many understand that Sonnet and even Haiku can probably accomplish their task, instead of them invoking a beast like Opus to tell them about todays weather.
And yet, MiMo and DeepSeek, even MiniMax, are way cheaper and arguably better, or way better than both Sonnet and especially Haiku.
While you can argue you are ready to pay 100-1000 times the price for Fable or Opus because you need those last 1-2% of edge, there's no valid reason to keep paying the obscene amounts of money for Sonnet and Haiku when alternatives exist.
I think that the reason is that American labs only focus on their biggest models, and make the smaller ones afterthoughts distilled from the main models. I wonder what would happen if they decided to optimize and train better smaller models.
The problem is, they can't. If they offer smaller models for less money, it would eat up their premium offers.
The only way forward for American companies is to push for always bigger and more expensive models while using regulatory pressure and fear mongering (FUD) to discard price effective competition from China.
I don't known how Codex works, but we can set environment variables and point Claude CLI to deepseek. I think that before slashing prices they will slash those environment variables. After all they are not working to give a free TUI to deepseek and possibly to other competitors. But eventually yes, prices will go down or there will be an attempt at a regulatory capture.
Honest question: what's garbage in that tool? I like that I can pair it with whatever editor I am using even if for 99% of the time I run it from inside emacs. I usually chat in the Claude window and it edits my files. I can also write a comment in the code, press a key and emacs will tell Claude to implement that comment. Simple needs, simple tool.
Most importantly, we need a model that doesn't randomly refuse us when we ask it to do something, or worse, deliberately sabotages us when it thinks we're building competing products. Like Anthropic's Fable.
I'm a fool for thinking that MiMo, in the context of Xiaomi who makes WiFi equipment (smartphones and routers), would be about network technology to manage parallel data streams (multiple inputs, multiple outputs https://en.wikipedia.org/wiki/MIMO)
I've worked a lot with MiMo in my project that pits LLMs against each other in games (clankerfights.ai). It is a very very good model for the price. MiniMax I'd say is smarter, but MiMo really touches near pareto frontier.
I found it relevant and actually just the information I was looking for. Having a highly recommended model behind the tool makes it worth further investigation.
Saying "This" on a post about "Mimo Code" leads one to, rightly, assume you are referring to "Mimo Code". If they commentor wanted to make a comment about a specific model it would have be clearer to simply mention the model instead of getting there transitively.
Good timing, I was looking for alternatives earlier today. opencode didn't install properly and I wasn't a fan of oh-my-pi and nanocoder.
MiMo code (via my z.ai coding plan) is very pleasant so far, nice UI and seems to respond faster than Claude Code. It might be injecting much less cruft into the conversation.
I also got access to the mimo-2.5-pro ultraspeed model yesterday, which is really quite snappy. It does cost more than DeepSeek, though, so I'm not sure whether it's worth it yet. Definitely fast though.
It did install, but then would hang when trying to configure the provider (specifically I was trying diffusiongemma via Nvidia). I gave up after a couple of attempts and moved to the next one.
it does have telemetry, enabled by default, that sends metrics to tracking.miui.com, including what model you are using. it can be turned off by environment variable (MIMOCODE_ENABLE_ANALYSIS=false), and yes it still has all the normal OpenCode provider logic so it will work with other/local models. it also automatically looks for updates and fetches a mimo model list, including when the telemetry is off, though those can also be disabled.
telemetry enabled by default and named "analysis" is not great.
Because they want to optimize it for their models and don't want to be blocked by waiting for PRs to merge or be rejected.
There's plenty of reasons to start your own fork that you have full agency of, as long as the OSS License is maintained anyone will be able to benefit from any new features they want to make use of.
KHTML is dead now, though. It was basically embraced, extended and extinguished by Apple and Google, who both wanted to take away the leverage of the community.
Today, legacy KHTML maintainers are boxed-out of upstream decisions that might prevent Manifest v2 from swirling down the drain. I'd argue the story isn't very beautiful anymore.
There are no X11 maintainers, there’s nobody to block out. In terms of users, of course there are X11 users. But nobody is maintaining it in any meaningful way and it’s been that way for years.
Also sysvinit is straight up obsolete software. Even if you hate systemd, there’s just no reason to use sysvinit because things like OpenRC and runit exist.
In addition, the adoption of systemd has lead to needing to maintain less init scripts because systemd units are just more portable. It used to be that every single distro and even versions of that distro required specialized init scripts for every application.
KDE is not a fork, nevertheless KDE was building on X11 heavily and now they are trying to kill it.
So while technically not the same story as if it was a direct fork of X11, practically they have the 'obligation' to support X11 indefinitely. KDE trying to kill X11 too is the same exact EEE that you mentioned as a sad story in regads of KHTML.
To go a different path perhaps? You can't expect that all your ideas will land into a main repo and you really want to implement your vision while using a sane base.
Because its currently impossible to "contribute to Opencode".
There are over 500 pages of open issues, up from 78 less than a month ago. They are doing nothing to halt the garbage/duplicates that pop up, and not even addressing legitimate PRs/reports.
Could just be a courtesy - Americans tend to be rather suspicious and hostile to contributions coming from China, and it might draw unwarranted attention from agencies and bad media.
Unlimited context catches m :D I understand that it's not possible but it's worth to be checked to understand how you do your context compressions, etc.
I'm pretty sure most of the comments in this thread are AI bots propping up this product?
I tried the free model and it's nowhere near Sonnet 4.6 in terms of capabilities. The fact that token speed will randomly get stuck at 0/s makes sense given it's a free service, but the way it performs is more reminiscent of AI from 2025.
It's "just" an opencode fork but it adds some nice features to try out while not being a full orchestrator metapackage like oh-my-opencode. Quite nice! Though it would be even nicer if this stuff came upstream or as an easy extension instead in the future
> MiMo Code is a terminal-based coding agent built by Xiaomi's MiMo team on top of OpenCode and open-sourced under the MIT license.
I think it is great that they built it on top of open code. Open Code harness is good and I want it to grow. Harness is very important and more projects use it, the more it is adopted.
My coding agent VT Code has recently become a Xiaomi Orbit partner. If you want to try out Xiaomi Mimo V2.5 and V2.5 Pro in a different harness, feel free to use my VT Code. VT Code supports Mimo V2.5/Pro via official Xiaomi endpoints and via OpenRouter. Thank you!
As much as I absolutely love Mimo V2.5 Pro (it's a genuinely good model), I absolutely hate the way they calculate usage in their token plan.
For example: For a super small task in a small project that should not be consuming more than 500K total tokens after all tool calls included, their shown usage shot up to 152 million tokens.
But, when I scroll down on the same page, a table shows usage as 3 million tokens, out of which 2.5 million were cached.
This is such a huge conflict on the very same page. The bad thing is that the usage progress bar is shown against that 150 million token usage, not against that 3 million one.
This has been in discussions for at least past 3 months on reddit as well, and was precisely the reason I subscribed to their lowest tier, and for a single month only.
Update: their own harness, mimocode, shows total token usage as just 63.1K. We now have 3 entirely different values, differing in 3 orders of magnitude.
Update 2: So, I did the exact same task this time using DS4Pro, and total token usage was just 101K (as shown by opencode).
Only tangentially related: MiMo-2.5Pro is fast, cheap and very capable, although not quite gpt5.5 level iontelligence (I dont use the claudes). It works flawlessly in Pi and is an excellent workhorse. I expect big things from their next model.
This is usually a PoC (Proof of concept) way to install something on a temporary container or temporary VM, but not for production use during daily desktop operation.
I was hoping their documentation would provide better installation instructions. But strangely, only for Windows do they recommend "npm install -g @mimo-ai/cli," which is a much better approach to managing installed packages.
For Mac/Linux, they have the strange recommendation to use the dangerous "curl <some_url> | bash." Quote:
> (for the best experience, Mac users are strongly encouraged to use iTerm or the VSCode Terminal)
> curl -fsSL https://mimo.xiaomi.com/install | bash
This is how everyone does it now. Including Anthropic.
To be fair, is that any different from naively trusting NPM? It's not like NPM is doing any vetting. They're every threat actors favorite sandbox these days.
You're right that it's as dangerous as it's executing random third-party code on your machine, but the method also has propagated far beyond PoCs and such at this point. All of these projects and many others push that install method: Bun, Deno, rustup, k3s, Docker (if using their helper script), Homebrew, Tailscale...
Frankly, it's not really more insecure than any other installation method. Apt packages and the like generally have the ability to specify pre/post-install scripts, so `sudo dpkg -i ./random.deb` is equivalent to `sudo bash ./random.sh`. Even if they didn't have pre/post-install scripts, they're still writing arbitrary files to arbitrary locations on your disk, so they can trigger execution the next time you boot or log in or whatever.
And at the end of the day, no matter the installation method (even just unpacking a tarball and executing the program directly from that directory), you're going to run their program on your computer, and then the program can do whatever it wants. Maybe you don't run it with sudo, but https://xkcd.com/1200/ seems relevant.
A package (like a .deb) is a static artifact. It can be hashed, mirrored, and GPG-signed. Package managers usually verify that signature before any pre/post-install scripts. A "curl <some_url> | bash" pipe is a dynamic stream; the server can perform targeted attacks: sending a clean script to 99% of users and a malicious payload only to a specific IP address or User-Agent. This allows for targeted attacks that are invisible to the rest of the community.
Yes, running third-party code is always a leap of faith, but why choose a delivery method that removes the possibility of verification and opens the door to targeted injections? Convenience shouldn't be an excuse to ignore basic security hygiene.
The problem is that npm, cargo, etc. set the standard in people's minds for how package managers work, when the Linux community has been working on securing the supply chain issues for decades.
Like requiring a WoT (usually with physical meetups) vetting people creating packages, FTP-masters, dedicated clean buildbots, etc. in addition to the packages themselves being signed and so on.
We've had this discussion since Eazel Linux desktop popularized bash | curl in 2001.
> npm install ... is a much better approach to managing installed packages.
No. Until the upcoming version of npm is out, npm will also run arbitrary code. Almost all common installation tools run arbitrary code. Not doing that is sadly the exception for now.
I get what you mean, but an NPM package is just a tarball of arbitrary code and some metadata. The whole point of it is to eventually run that arbitrary code, presumably. Otherwise why would you want to download the tarball and extract it? In fact, what purpose does NPM even serve if it's just a way to host tarballs?
I get the install time and run time execution might feel different, but I don't see how that's a security boundary at all.
I suspect that everyone will just get into the habit of typing --allowScripts all or whatever and nothing will actually change, because there's no point in a version of NPM that doesn't properly set things up for most people.
The code in the module isn’t arbitrary: it’s what the user intended an install, provided the package hasn’t been compromised. I do see your own version though.
Most apps don’t need install scripts so disabling them by default is fine.
Isn't Unlimited Context pretty difficult to promise? What exactly do they mean, could I just have two agents locked into a TTRPG back and forth forever?
You can run the two agents in GAN like loop.. each trying to better the other. Give them a common task like design a better alternative to transformer model that uses max O(nlogn) memory, and the result comes closest to existing n^n implementations.
Good idea actually.. why haven't I tried this before.
Terminals and things that live in terminals have relied on wcwidth() to handle this since time immemorial (which is always fun when they are out of sync, e. g. remote over ssh, but in the vast majority of the cases it just works).
Strange, I reckon I installed Tahoe just a while ago and still didn't have a similar issue, but I remember on previous MacOS versions the error message for unotarized binaries, was to warn that there was indeed a security issue with the binary, not that it was simply "damaged".
Reading "xiaomi"in the headline I was thrilled to see MiMo only to find out it's NOT this one: https://en.wikipedia.org/wiki/MIMO
What a disappointment!
Let the battle for the harnesses give free tokens for all, until the next competition arena does the same. It seems that's the only way AI will remain acessible.
This website is gorgeous, by the way. The mouse reveal on the background, amazing.
Is that Open-Source like, run it locally, no phone home included, or open source like the thin front-end layer is all that is actually open-source but it’s an empty shell without the remote API it relies on?
They default it to talking to a free version of their model (which is incredibly cheap if you decide you like it.)
But it seems trivially easy to run it against local models. Their onboarding guide offers that option, though I have no idea if it changes any functionality.
The latter. It looks like it's meant to be a batteries-included agent to promote their free-for-a-limited time AI service that it connects to by default.
Ok, fair enough compared to the rest of the proeminent actors I guess, but quite confusing from dev point of view. Lately I started to experiment with model like Qwen2.5 on local. Good enough to ask simple question, but didn’t manage to do anything remotely close a agents I started to experiment with through Copilot.
sounds really cool if it was coming out of anywhere except China, which has laws to exfiltrate your data and send it back to the government for espionage purposes [0].
Today on open source vampires: xaomi forks an open source project, doesn't contribute upstream, attaches usage restrictions that are probably incompatible with the license, and wants good PR. Fuck these people.
Open source and open weight AI is very important to protect freedom of speech. OpenAI, and ESPECIALLY Anthropic, will try to ban them through regulatory capture / safety fearmongering. We need to make sure that does not happen. It’s not society’s problem if these frontier labs have no moats.
Good, coding harnesses should be open source and LLMs should be treated as commodities. Minimize switching costs for consumers, and let people understand how they're interacting with the context and the LLM outputs.
The industry has been moving the wrong direction with Claude Code staying closed (despite multiple times leaking the source code!) and the open source Gemini CLI being deprecated in favor of closed source Antigravity CLI.
Why would a company do any of these things? What is their motivation for any of it? That’s like saying cloud providers should be commodity and should open source all of their platforms and eliminate egress fees so customers can easily leave at any point in time.
That’s a charity, not a business model.
On the question of language models and periphery tooling -
Open weight models are disruptive to the business models of closed model businesses. An incentive is if your business is built around X but model training is helpful to you, but you don’t expect to meter it specifically. You can release your models and undercut the exclusive moat of a new model company like OpenAI or Anthropic from becoming at some point a competitor, or holding their access as a chip in pricing negotiations. By opening your architectures and weights other competitors can build on them and newer better models emerge faster decoupled from a small number of proprietary models. This lets you focus on X while gaining overall momentum on your model release at no additional cost and no loss in focus on X, while defending against upstarts and monopolies.
This is effectively a lot of the open source world that comes from corporate development as well. It feels odd after this many decades of discussing corporate reasons to participate in open source we keep rehashing it.
Maybe these things should be utilities that can be swapped out at will and shouldn’t even be privately owned at all? Heresy, I know!
Having a five-year plan dictating which publicly owned b2b SaaS AI service you will be provisioned sure sounds like a dream of mine. I wonder what could go wrong.
Open source is as opposite as state ownership as it is to private ownership.
Open Source comes out of Anarchist philosophy not Communist State owned type ideas. There is no direct ownership, its a project built as a common good, and owned by "the public". Its a community garden that people are free to plant and harvest as they see fit.
The concept of a common-pool resource has nothing to do with awkward radical political ideologies.
Yes I do think cloud providers should open source all of their platforms, and this is not charity because it is essentially the hosting that they are providing as a product. Even if, say, google open sources its whole search infrastructure, it does not at all means you can just host your own due to the huge hardware requirements, but you can know(especially after AI which can be utilized to do this) that they are not using your data in a way they shouldnt.
> because it is essentially the hosting that they are providing as a product
That would be true _if_ they were forced to open source all their code, but it isn’t today.
> Even if, say, google open sources its whole search infrastructure, it does not at all means you can just host your own due to the huge hardware requirements
You can’t but Acme Inc and other bigcorps could, and Google’s margins would evaporate overnight.
Because there is literally nothing special about coding hardnesses. The models are doing all the lifting. It just user experience that separates them.
A coding hardness with just bash outperforms Codex, Claude Code, OpenCode, Pi ect. The added features are just user experience features.
If harnesses are basically doing nothing, why would these metrics vary so widely?
https://www.endorlabs.com/research/ai-code-security-benchmar...
There's a lot of ways to configure agents and any implicit configuration to harnesses may have a non-trivial effect.
It's because they do things that is why they score differently. Coding hardness add features for user experience not for agent efficiency. If they did all the coding hardnesses would be using bash and code mode and letting the agents write code to perform tasks but this doesn't work because you want humans in the loop. You want users to be able to approve and deny writes. You want uses to see edits. So you have to build tool for these. It's hard to show diffs when the agent is just using bash.
> The added features are just user experience features.
> It's because they do things that is why they score differently.
That was my point. Regardless of how you feel about UX, it's a value added set of features. The question initially posited, stands. Why would a company do any of these things?
> Coding hardness add features for user experience not for agent efficiency.
Pretending it was always about some metric you just decided was important is moving the goalpost. It's not compelling.
I think it makes more sense that it's Freemium Dominance or they act as Low-Cost Marketing tools.
Try Kimi in Kimi CLI and Claude Code and try saying that again. Kimi quickly collapses into tool calling loops without measures in their CLI but not in Claude Code and is largely useless for any long running tasks in harnesses not taking this into account.
With those measures (which are actually quite interesting) it can at times perform at Sonnet level.
A harness(notice the lack of a 'd') is a strap system to gain control over something.
Like the thing people attach a dog lead to so that their kids won't just go kamikaze into a car.
Coding harnesses are named by analogy to that.
They are not hard.
The reason I have a dog harness is to distributes weight so I don't choke her when she goes at the other dog that she doesn't like. I'm actually puzzling over kids kamikazeing into cars
It is a common fear for parents. Obviously they are not fighting for the emperor but chasing or running away from something.
The strapped kids are often normal with no apparent disabilities(but it is possible they have an ADHD diagnosis).
Never thought about doing it to my own.
It's actually only a problem if it's the other way around, isn't it?
If kids run into a car, they will most probably just bounce and continue, perhaps inflicting some minor damage. But if a car mows down a kid, that could well be a fatal injury. Leashes for all the cars! ;)
A kid running into the side of a car that’s moving over a certain speed will be lucky to break only a few bones.
You got to miss spell these days or people assume your ai :)
That's very punnyy
Its like yuo're on fire!
Your reply doesn't answer the question: What is their motivation for any of it?
I would disagree here.
Building a good and working coding harness with smaller models is really hard. Everything evolves around the limited context size.
Tools must be specification driven to reduce noise and high temp hallucinations, tool call shrinking needs to remove errors and tryouts of different formats of parameters (because LLMs always ignore descriptions in the JSON...), and you have to deal with long running agents because you can't afford them. Planner/orchestrator architecture, agent to agent communication need to be summarized, and then you have the messed up scheduling parts, because you need to prioritize short running agents and give the planner a tool to wait for outputs of spawned contractor agents.
And that's not even talking about sandbox vs playground read/write/access policies of tools.
Harness engineering, if done correctly, is quite hard.
And all of this works 60% of the time, every time.
Anyways, that was somewhat the summary of the last 6 months building my exocomp agentic environment. And it's still not satisfying to work with.
In my limited experience, the smaller the model, the bigger the harness. Where with something like claude or deepseek the context size etc just let's you give it bash access and step back; small models tends to do better with simple action - response , new context each call. Context management becomes a continuous activity. Its a fun space , and I have found big models decent at building and improving these harnesses for the small ones. Using /loop and just run a continuous test - build - test loop.
Because they steal everything to train their models. They literally make you pay for the "commons" knowledge
Cloud providers are commodity, and egress pricing is partially cost following because they have to pay peering to their interconnect points for WAN. Internal networks are not charged within the account because the economics of the VPC overlay are optimized for that use, but inter account and VPC and other boundaries carry cost - especially interconnection between accounts because the way VPC treats virtualization requires a relatively expensive routing. Inter AZ and inter region pricing also exists for the same cost following reasons. They also help shape incentives because it allows them to optimize placement of compute within the same AZ to physical buildings or rings.
The case that is largely nonsense is the egress pricing on direct connects since beyond the circuit costs, which the customer pay, there’s no costs for aws not already on the customer regardless. It also makes DC friction weird in that you are incentivized to NOT move storage before compute.
At least AWS egress pricing is a very high multiple of market rate.
opensourcing software may enable leverage of wider network of contributors to given piece of software,hence software can evolve much more quickly and efficiently.
there is no proof of that whatsoever.
the internet is full of semi skilled "developers" who should never be developers in the first place.
> That’s a charity, not a business model.
https://gwern.net/complementHere's the Spolsky article, called "Strategy Letter V": https://www.joelonsoftware.com/2002/06/12/strategy-letter-v/
Here's links to the whole series up to VI: https://lettersremain.com/joel-spolsky-on-strategy/
"Commoditizing your complement" is a convenient market phenomenon when it happens, but it's not a hard rule that everything that should be a commodity lands in the window of incentives that result in this strategy being the self-serving optimum for a company. Usage of "should" here can be interpreted as "would be most beneficial to the largest number of people in terms of cumulative productivity" or "not having this be commoditized leads to terrible incentives that harm a lot of people." There's no guarantee at all that it ends up being most profitable to individual market actors to attempt to commodify LLM compute, and in fact they seem more interested in securing legal moats and protected monopolies. You may think that reality will force this to fail via open source model releases from China or the like, but consider the world in which the US officially bans import and usage of all but specifically licensed and approved LLMs. The most powerful companies and people here seem broadly aligned on wanting that, so it's not implausible. Sure, you as an individual could illegally import and run an unsanctioned LLM yourself, but most people won't do that, and attempts to scale it would be beaten down with the force of law. The companies who strike deals will get 99+% of business and profit handsomely, no need to ever let their golden goose be commodified. It's a strictly better strategy if your government is dysfunctional enough to let it happen.
> but consider the world in which the US officially bans import and usage of all but specifically licensed and approved LLMs.
Then the US will fall behind China and, more importantly, from the rest of the countries in the world who refuse to lobotomize themselves in the manner you described.
There are also other cultural and political risks which, although delayed, may easily grow into something apocalyptic. Rich people building bunkers isn't just a fad, the outcome they envision isn't fiction, it might actually be a part of the plan, and while this is concerning it does make everything simpler.
I don't really see how this applies to Anthropic. Claude Code _is_ their integral money maker. Their point above is a three-way neck-to-neck competition where they won't be able to become a near-monopoly any time soon.
> it can be more valuable to make it free if it increases profits elsewhere.
That's a big "if" for Claude Code, et al.
Is it? Seems from the outside that it was the thing that finally drove Anthropic into profitability.
What drove them to profitability?
Anthropic was making moderate revenue from their API tokens, but everything changed when they started bundling crazy amounts of usage credits at $20/$100/$200 claude max subscriptions relative to the API token rate, but only as long as you use their coding harness — despite claude code, back when this started at least, being an inferior coding harness compared to many open source ones.
CC is technically a free product (you can use it with any model). It's got very few popular "opinions" on how a coding harness should work, but it's by far the most popular one (including at competing LLM manufactories like OpenAI and Meta and Google). Why? If it was just that their models are 5% better, most workplaces would optimize on token (aka cost) efficiency.
Anthropic has been winning the usage of their harness, their tokens, while earning significant revenue, by significantly subsiding their token consumption.
This has earned them many things:
- prime data on how software development — simultaneously the leading beneficiary industry from LLM use, and also the most flush with cash to spend — has been using LLMs
- bringing that industry to standardize around their harness concepts. They are essentially establishing themselves as the W3C of LLM interfacing, except as a private organization.
- all dat data
> cloud providers should be commodity
They are already.
> and should open source all of their platforms
Most of the cloud platforms are open source. Linux, container, k8s… it’s entirely possible for someone to build and deploy their private cloud if they have the resources.
> and eliminate egress fees
What does it mean? If I sign up for cloud service I am only bound to the contract terms. If I am PAYGO I can switch anytime.
> Most of the cloud platforms are open source. Linux, container, k8s…
Linux isn't a cloud platform and neither is Docker.
Kubernetes was created specifically to create a way in against AWS' de facto public cloud monopoly. The Cloud Native Compute Foundation is a classic "alliance of smaller players uses open-source and interoperability as a wedge against an incumbent that threatens hegemony".
> it’s entirely possible for someone to build and deploy their private cloud if they have the resources.
What is there for that, really? Basically just OpenStack?
I never say these pieces are the equivalent of a cloud platform. What I said is that with them, it is entirely possible for someone to build their own.
Why someone wants to do that is not my concern. But not being able to is.
The cloud provider isn't the harness, Terraform/OpenTofu/Pelumi and the abstractions you build using them are. The cloud provider is the LLM. It's not as fungible as the LLM and there's no direct comparison to egress costs of course, but that's moreso a problem with the metaphor.
Because Claude Code is literally nothing particularly special. We don't need their business model. They need their business model, and to that I say, tough shit.
Well, until you established monopoly you need to build trust. Open Source is one way of doing just that. One of the better ways I would say even...
Public good isn’t a charity, and a business model that doesn’t contribute to the public good should not be allowed to exist.
Who defines what is public good?
Is it public good? Or is that a coincidence covering the real motive of an attempt to undermine the viability of American companies?
What?
China is giving away AI for free so it’s harder to make money. The same strategy they did with solar panels. Sell them at a loss long enough until the manufacturers go out of business and you’re the only one surviving. Then flip to extract monopoly profits
What does it sound so familiar with some US business models? I mean all the big tech does exactly that plus Uber (if you call it a tech company)
Instead we got home solar system become very affordable over the past decade.
And driving out US manufacturers isn’t even the main goal for China. They know their huge risk on reliance on petroleum and was doing everything they can to mitigate that. Building out a huge solar manufacturing base is their answer. Now they are reducing petro imports YoY.
> China is giving away AI for free so it’s harder to make money.
That might've crossed their minds but that wouldn't move their hand, not even a finger. Politics is the primary driver here, here's the deal:
AI is the new Internet
China foresaw a world where they'd be blocked from it and Anthropic's ongoing attempt to block their own country form it shows how right the Chinese were.
Does that apply to US open weights models too? Was Llama an attempt by Meta to destroy America's economy?
Yes. Releasing open weights seems to be an effort to slow down competition. Facebook was behind. Grok was behind.
Don’t get me wrong, I am glad they are doing it, I personally use open source models. In part to not spend money on the other APIs. So it’s clear that some percentage of what would be paying uses choose to not pay. The other part of opensource is free community labor. How much development work did Facebook get for free around the core infrastructure of react by open sourcing it? A massive amount that they did pay for but benefit from.
The point is, it may benefit society, and yet that social benefit wasn’t the motive for them releasing it as open source.
I don’t think the motive matters to a large extent if the effect is good.
Motive probabaly doesn't matter in the end, outcome does. But understanding the motive is a good thing.
Isn’t that the standard economic playbook for all well funded businesses at this point? Walmart, Uber, Amazon, etc
Do you think Internet Explorer 6.0 was a good decision?
There is a difference between a loss leader that drives a monopoly. Microsoft gave away a browser to make their monopoly OS more valuable and deepen their anti competitive moat. They were almost broken up as a result and honestly should have been if we had any anti trust enforcement.
How does open sourcing a Claude code clone drive adoption of anything that is a monopoly or even commercially related? Instead it seems like an attempt to undermine US AI companies.
That being said I am increasingly skeptical of how the US leaders are converging on creating monopolies and going deeper into the app layer that means they will end up owning everything rather than being the substrate of a competitive and flourishing ecosystem.
If it were up to me I would implement a regulation that 1) AI labs can’t own inference hardware, data centers would be a regulated utility like electricity and the internet required to provide open access third party safety, guardrails and audit, 2) inference providers can’t build apps beyond serving API requests 3) training data sets are required to be open sourced within 3 years of training a model.
What we are doing now is allowing vertical and horizontal integration of the hardware, training, inference and application layer. Last time we did that standard oil ended up owning the rail network, pipelines, oil fields, gas stations and refineries. Go see how that worked out for society.
They did steal all of our written knowledge
Good will and trust can ultimately have monetary value, and having a funnel based on open source is a viable play if it leads to a service that is sticky.
> Why would a company do any of these things?
What? It’s actually insane that they haven’t yet.
I don’t like changing tools. What engineer does? I want to learn one tool and tune it to my exact preferences. Proprietary vendor tools are not portable and I avoid them.
Either Anthropic or OpenAI could drop the first-to-market open coding harness tomorrow and it would be as big as VSCode, it would be the standard platform everyone builds stuff on.
MS releasing VS code led to Cursor reaping $60B in value.
The familiar Chinese recipe for success: Always copy and imitate first, even if it is inferior, always make it cheaper or even free so that the original innovator will be burdened by brutal price competition and much bigger R&D costs and cannot keep up in the long run. Then the copycat will win in the endgame.
There are very rarely original ideas or original products. It’s more about money put into that idea and the marketing. You could say that Apple copies and imitates just like the chinese. After all they wait for the others to pour money into the ideas and they just deliver a polished experience on top of the hard work others did. Yet people don’t accuse them of being imitators.I think people are mad at the Chinese because they do it for cheap and at an industrial scale on so many products/industries(i.e even food). To me they are a force for good competition.
did you just time travelled from 2010?
show me from which country or company did the Chinese copied their EV techs, batteries, drones, robotics etc. for their military gears, show me show they copied their J-36 and J-50 fighter jets, Type-055 destroyers, YJ-21 missile and the most recent 09x sub.
time to stop living your socially isolated life and start reading stuff other than whatever fed to you by your brainwashing media.
The capital motivation isn't the only one that exists. You can say something should be true without having a plan to maximize quarterly revenues.
Even if you consider profit motive, what is the profit motive for corporate contributions to open source? The same applies here.
“A business that does things that customers actually like is a charity” lmao
This ^
After 16 years on Hacker News, I've come to associate its readership with cheap bastards who think everything should be free while simultaneously wanting to keep their 6-figure jobs.
There's a very strong overlap with male gamers, who also think everything involving sophisticated engineering and design should be cheaper than a cup of coffee.
Just call it out and maybe we can collectively choose to towards a culture that doesn't encourage such shameless behavior or perverted values.
> on Hacker News, I've come to associate its readership with cheap bastards
They might be cheap but they surely aren't stupid. They sure know who not to trust, security beats imaginary profits promised by cheap used-car salesmen who did move software jobs to wherever they could spare a dime.
And you want us to defend their current profits which they swear to use for the removal of the remaining software jobs...
If you see this as people just want free stuffs you are missing the point. There is an entire different level of danger to allow a few entities to completely own a day to day technology.
Imagine personal computers and open source operating systems completely don’t exist. What you can do on a pc today, you have to always do it on a thin client always connected to a main frame run by a selected few. Everything you do is recorded and subjected to at least 30 days of retention and inspection.
Imagine every car is completely not customer serviceable and have to be connected to one of the three manufacturers in order to operate. Everywhere you go is recorded. The manufacturer may decided the place you want to go is inappropriate for you to go.
You’re not totally wrong but I think the motivations are varied. Using myself as an example, I have paid huge sums for software, films, news, and yes open source software, but I hate paying 5 bucks for a sub or buying certain products because the deal just isn’t a fair trade. I’d say most modern software is a bad deal.
That said I am a cheap bastard.
I'd gladly pay for software that demonstrates craftsmanship, quality and care for the user. Not a very common phenomenon unfortunately.
Note it is easy to confuse "free" with "paid by advertisements".
The jagged frontier of frontier models means treating tokens as fungible between providers is naive at the limit of capability, but also will work for solved problems far from the boundary. The problem is you need to keep evaluating all models to know where your use case lies on the frontier map.
As a concrete example, you’ll get very different results for the same prompt for sonnet, opus, fable, gemini, gpt 5.5, …
ah nevermind it's just a fork of OpenCode
There is really nothing free in terms of money, there are only things really free in term of spirit. But AI coding assistant are not those things related to spiritual freedom.
The complements ARE the LLM AND the harness. The actual products we're all consuming are GPUs. Memory being expensive is a second-order effect.
The platform is the GPU, and doing cool shit with it IS the complement, which requires more memory. And demand is so high and will stay high, that it looks like the platform itself.
> And demand is so high and will stay high,
The question is why supply is restricted, primarily by sanctions and tariffs to China, and the expressed refusal of RAM makers to even think about increasing supply, they are actually all sweaty about China taking a bit of the unrestricted market.
what do you mean by should?
As much as their propaganda wants us to believe, Anthropic is not 'the industry'.
> MiMoCode is built as a fork of OpenCode. It keeps all core OpenCode capabilities (multiple providers, TUI, LSP, MCP, plugins) and adds persistent memory, intelligent context management, subagent orchestration, goal-driven autonomous loops, compose workflows, and self-improvement via dream/distill.
From github
Sounds like they slapped in a bunch of common plugins and released it as a product to promote the free-for-a-limited-time use of their new coding AI service.
> promote the free-for-a-limited-time use of their new coding AI service
Not sure which "free" service you're referring to, but MiMo v2.5 Pro is plenty capable & (after its recent 70%+ price drop) one of the most affordable options in its class (DeepSeek v4 Pro, MiniMax M3, & Qwen 3.7 Plus). I read somewhere that Labs are incentivized to implement custom harnesses because each model has its strengths, quirks, & blindspots (like Qwen forking Gemini CLI)?
I'm guessing the greatest reason behind each provider creating and agent harness is that (a) there is not a clear winner still and (b) it is harder to switch models with a competitor, as you also have to switch harness
But you really don't have to switch. MiMo Code has the same provider support as OpenCode.
Even Claude Code you can use with any provider that exposes an anthropic API endpoint, which they all do.
Or by using a proxy, yeah. Personally I would still prefer a multi provider harness over CC when using it with another provider, if alone for the visible reasoning, model switcher, cost estimation and so on. So far I've only preferred CC when I needed to work with Jupyter Notebooks because it has built-in tools for that.
> like Qwen forking Gemini CLI
That was a good call. Gemini CLI is dead.
don't anthropomorphize the guillotine
What I like about MiMo too is that it is multimodal.
For example, I can send screenshots of what I'm developing and it understands.
Only non-pro is multimodal I believe
I thought only the MmiMo 2.5 non-pro was multimodal?
So, basically the same thing silicon valley has been doing for the past half decade.
Since the link is in Chinese: MiMo Code is Xiaomi’s AI agentic coding harness.
“ MiMoCode is a terminal-native AI coding assistant. It can read and write code, run commands, manage Git, and use a persistent memory system to keep a deep understanding of your project across sessions while continuously improving itself.”
GitHub link (English): https://github.com/XiaomiMiMo/MiMo-Code
@dang might be better to link to the GitHub, and not for language reasons.
(Edit: for posterity, original URL as submitted was [0]).
[0]: https://mimo.xiaomi.com/mimocode
You can change the language via the header: The rightmost option is a language dropdown.
It's a client-side change and doesn't impact the URL so users must manually change it each time they visit the site though
Thanks, I missed that on first glance and did manual translation.
Not sure why my iPhone shows an option to translate website but all the destination languages to pick from (I have multiple languages installed), including English, are greyed out. iPhone does support translating from Chinese (Simplified or Traditional), and the button to translate website isn’t greyed out like it is for unsupported/unrecognized languages. Might be an iOS 27 bug, because it is working on other websites?
It's entirely possible, and even standard, to allow the browser to tell your site which language to respond in.
While ignorance of internationalization standards is a possibility, and the most likely cause.. I do wonder if it's a bit of a nudge to promote Chinese influence in the AI space.
Not that they really need to do that, China is already doing great (relatively, depending on criteria). The implosion of the US, the resulting brain drain and world shake-up has been very timely for their AI and other industries.
It's a very smart move for them to think longer term and start freezing out NVIDIA. Then they can take Taiwan purely for ideological concerns and not worry at all about the fabs blowing up in the process.
And they won't be dependent on foreign factories sitting on an island just off the shore of a superpower who's shown nothing below absolute resolve for decades towards the idea of conquering that island....
Why not persist it through a query param? Or a lang param for that matter
Feels like maybe you're just noticing this because it defaults to Chinese. Is that true?
How many sites do this but you don't notice because they default to English?
You're right, there are probably lots of sites misconfigured to not respect language headers, but we don't notice because English is the default.
However, the right solution is still to use the language header. I send that to them, they should use it to give me the right one by default.
One of the funny things is that this whole site is in an iframe; which breaks both Google Translate, and the Firefox translate feature. If you check, the outer iframe seems to indicate `lang="en'` and loads the iframe with `src="/coder/index.html?lang=en"`, but the inner iframe still gets a `lang="zh-CN'` by default until you use the toggle.
If you go to the eventual redirect source of the page with `lang=en` parameter, you get a `lang="en"` attribute, but it's still in Chinese until you toggle it with the menu: https://mimo.xiaomi.com/coder?lang=en
Anyhow, yeah, lots of pages are probably broken this way but we don't notice. But still, it has that info from your request, it should use it.
But what if you have English configured as a preferred language? Isn't that what it's for? Wouldn't it make sense for a website to respect that (when available)? I hate that google.com doesn't and defaults to random languages based on IP.
Web standard often give great grounds to leverage on. Modern stacks often really poorly work with a lot its surface and reinvente half baked bespoke alternatives.
Personally, none, I’m not English native. I didn’t notice the locale switch, but mostly because the look of the website was so beautiful I didn’t pay attention to menus. I wonder if ideograms keeps looking so beautiful once you learned to decode them. I never found Latin script to be particularly beautiful, and to this date Arabic script remains my favorite one in term of esthetic (I can’t read Arabic ever).
This is not exactly what you’re asking about, but I started learning Japanese when I was in the middle of playing Cyberpunk 2077 (for unrelated reasons); and I gotta tell you; realizing that 98% of the Japanese text everywhere in the game was just “hotel” or “karaoke” definitely took away some charm from it.
This is really strange, but the link provided has an iframe pointing to https://mimo.xiaomi.com/coder
When you look at the source, you can set it to English from the params https://mimo.xiaomi.com/coder?lang=en but there's a small bug, the hero subtitle isn't translated but everything else is.
Wondering why this is in an iframe.... so strange
Language support was probably an afterthought since their target audience all read Chinese
What a transformation by Xiaomi to build almost frontier level models. Five years back, when I was in the data science team, they dint really bother about AI models and were using Baidu for NLP and vision under the hood of their APIs
I remember when they made this:
https://en.wikipedia.org/wiki/Xiaomi_Mi_1
And now they make one of the fastest cars ever created and frontier-level AI. In just over a decade. 你好!
Wrote this eons ago:
https://news.ycombinator.com/item?id=9421471"Death of Silicon Valley" in this case is such a funny perspective. Like, how twisted is the US's view of the market that they think "Competition? Oh no. Sound the alarms."
Except it’s not competition if US companies can’t access the Chinese market but Chinese companies can access the US market. Just like cars. America is not willing to compete with BYD. But we are 20 years into massive IP theft from China and the naive and short sided leadership in the US that basically traded our knowledge, design and manufacturing knowledge for cheap of shoring, and watch China execute spectacularly to take advantage of the opportunity.
> short sided leadership in the US that basically traded our knowledge, design and manufacturing knowledge for cheap of shoring
The shareholders and execs got rich, you are not rewarded for very long term performance so why bother thinking what will happen 10 - 20 years later.
I mean, sure, that's just protectionism. It sounds like you're against it.
Xiaomi have been cooking a lot in recent times. Their model, especially the pro series, is underrated in my opinion. It haven't received the attention it deserves while it is pushing higher and higher in benchmark scores (looking at artifical analysis), and this was before Deepseek dropped V4.
Furthermore, their pricing plan is insanely cheap, they even upped usage limit for their cheapest plan, lite plan, which is at 5$ / month. And now, they are dropping a Harness for their own model? Amazing. I wish they added support for installation through Homebrew though.
On another note, this is what I would like to see more of from a company, what I do not welcome is startups making their model exclusive and hurt their customer base through sabotaging as a way to prevent eventual distillation attempts.
>Furthermore, their pricing plan is insanely cheap, they even upped usage limit for their cheapest plan, lite plan, which is at 5$ / month.
Unless something changed their plans aren't really worth getting. They're not that much cheaper than the per-token rates, and because it's a plan, you have to contend with weird usage restrictions. You're better off paying per-token unless you have some use case that demands a very steady stream of tokens.
Wow. I saw 4.1B credits and thought it was super generous. But my math says the subscription plan gives less value than the API.
For example, API input is $0.435 / M tokens, which works out to 13.79 M tokens for $6.
Plan is 300 credits per input token, which works out to 13.67 M tokens at 4.1B credits per $6.
Very similar math for cache input and output.
Looks like they have very effective collaboration with DeepSeek and Kimi. Those three models have been bouncing ideas and sharing R&D innovation, which made all of them improve very fast.
Based solely on quality and price, OpenAI, Anthropic, and other western models just can't compete with the new generation of Chinese open models.
>Looks like they have very effective collaboration with DeepSeek and Kimi.
The collaboration is informal. People don’t seem to realize this, but the Chinese internet for programmers and developers today feels a lot like StackExchange in its heyday. There’s a huge emphasis on sharing knowledge, because sharing what you know builds your profile, and becoming a rockstar in a subfield is one of the only ways to get ahead.
Competition in China is ruthless. But unlike in North America, where individuals are often bound by agreement to hoard knowledge because it can give them a competitive edge, the competitive advantage in China is building face and peer recognition. And that comes from proving that you are worthy of being a "master/teacher", and that extends to the valuation of your knowledge business. For example, the third wave coffee shops in China, the master roaster is often called "master/teacher" once they win a roasting competition and start sharing new knowledge of roasting in the public sphere, and that's a title of sincere respect.
You can see parallels with those that apply to give talks at conferences and post snazzy technical presentations they give in the US, but the bar for what qualifies as new knowledge is far higher in China because there's a massive ecosystem of people rushing to outcompete what you have to offer, and once the ball gets rolling on knowledge sharing, lots of people will go off and build upon that knowledge or try to build businesses on top of that, which in turn produces more knowledge.
Reading developer forums in China, once you crack the code (I find Gemini will get you a good chunk of the way with good translations), they are really quite far ahead with what they're willing to share. And I suspect in great part, the decision to release open-weights is heavily tied to that concept of building face/peer recognition = building valuations.
> huge emphasis on sharing knowledge
> developer forums in China
Care to share a few such forums or websites, genuinely curious as someone who reads basic Chinese.
A decade ago, I knew there were CSDN, jvjin (may not be the correct pinyin) where people talked about tech. Nowadays they have just become SEO farms like the rest of the internet.
Actually there are very few Chinese forums left, because it is a huge hustle to run one, and the same rule applies to the comment section of any personal blogs.
Right now pretty much the only one left with a bit discussion is v2ex, but it's blocked by gfw and kinda becoming a ranting corner for programmers. Some also recommend nodeseek and a couple other new sites, where you can find shady stuff but not tech stuff. Then you have tech blogs of big corps, or personal ones, but still not many quality tech posts. There are also some on WeChat but it's kinda cumbersome to use so I stopped.
Thank you for the insights, this sounds like an environment I want to partake.
> Reading developer forums in China, once you crack the code (I find Gemini will get you a good chunk of the way with good translations), they are really quite far ahead with what they're willing to share.
Are you able to share those forums and other resources? I would love to read what people in these communities are sharing.
What forums do Chinese programmers use that you would recommend exploring around in?
Mainly want to see the people who are building stuff I've never even thought of.
> Looks like they have very effective collaboration with DeepSeek and Kimi. Those three models have been bouncing ideas and sharing R&D innovation, which made all of them improve very fast.
Very fascinating to learn this, didn't know Moonshoot (Kimi) also collaborated with others. I think I read in another post that DeepSeek and Qwen team shared the same building? So that kind of explains it.
> Based solely on quality and price, OpenAI, Anthropic, and other western models just can't compete with the new generation of Chinese open models.
I have to agree. I had the great opportunity to take the offer Z.ai had with their Christmas deal, their lite plan was 3 months for 7$. GLM-4.7 was already impressive enough.
When they released GLM-5-Turbo and GLM-5.1, that is when I came to the realization of how close the gap is between proprietary western models and Chinese open-weight ones (not all of them are ofc).
I could barely believe how good GLM-5.1 was, I didn't think I was using it in CC and had to check the settings again. It's astonishing how close the gap is now, and this competition benefits us very much, the pricing is so low atm, its amazing.
Consider also Chiness EV prices. Unfortunately you can't buy them in the US it seems. https://www.reuters.com/business/autos-transportation/averag...
Pretty neat that you can just install it and start using it (at a Sonnet 4.6-level model) without needing to sign in or pay.
Typically, Chinese websites are a big pain to log in or sign up because they require a +86 phone number due to legal reasons. Being able to use it without having to make an account is amazing for friction reduction. I could probably even just install it onto new machines to help with set up.
I wonder how they are gonna detect and block abuse though?
> at a Sonnet 4.6-level model
MiMo v2.5.0-Pro is honestly the first Chinese model that I've tried where I really though why should I use Claude Sonnet when I can get the same results for a fraction of the cost. There was always something off about Chinese models that made it apparent that it couldn't fully compete with GPT, Claude, Gemini, etc. but this was the first model where I was like, this feels like Sonnet.
I can't prove it, but I think they trained heavily on Claude output. From my perspective I don't care since Anthropic trained on my data.
Using them also works well for North Americans as our peak hours is not theirs.
If I had one complaint, the v2.5.0-Pro model thinks too much.
I find deepseek-v4-pro to be every bit as good as sonnet tbh
Is there a guide to running these models locally? Sonnet level inference on my own hardware would be world changing.
I have Claude but I don't want to ask it because Anthropic could decide to sabotage me.
They won't be giving this away, at least not for some years. It almost certainly has distillation data embedded in it, and that would be a smoking gun.
What? I just searched the web and the results say MiMo V2.5 Pro is fully open source. The weights seem to be out there.
Distillation is not a problem.
There are certainly open weights which are available, I don't know if that is what is running on the service.
GLM 5.1 is stronger than Sonnet 4.6 in my opinion, but while they have a coding plan that is a good value MiMo beats it on price. I haven't used MiMo much yet but it felt pretty similar.
“Just install it and use it” is great UX right up until the first botnet also discovers great UX.
Xiaomi has been selling physical products to West through their websites for 10+ years now. I dont think you will encounter any issues here
So funny I have noticed how terrible the signup is on all these Chinese models, companies etc. Always wonder why it is such an easy process. Like QQ, Tencent etc demos Ive seen past year
Much more information in the blog post this links to: https://mimo.xiaomi.com/blog/mimo-code-long-horizon
Terrific link thanks for highlighting it
Claude and Codex pricing will eventually have to come down, for most common coding tasks you don't need a super smart slow model but a smart-enough and very fast one.
I’m not convinced they can come down, especially as they both are opening their books for S1 IPO filing.
cheap token for the win.
Microsoft github copilot recently changed their billing. i'm on the yearly subscription. GPT-5.4 is now 6x and even previously free model like GPT-5 mini now cost .33x. its only June 11 and my usage is now at 50%.
I don't think many understand that Sonnet and even Haiku can probably accomplish their task, instead of them invoking a beast like Opus to tell them about todays weather.
And yet, MiMo and DeepSeek, even MiniMax, are way cheaper and arguably better, or way better than both Sonnet and especially Haiku.
While you can argue you are ready to pay 100-1000 times the price for Fable or Opus because you need those last 1-2% of edge, there's no valid reason to keep paying the obscene amounts of money for Sonnet and Haiku when alternatives exist.
I think that the reason is that American labs only focus on their biggest models, and make the smaller ones afterthoughts distilled from the main models. I wonder what would happen if they decided to optimize and train better smaller models.
The problem is, they can't. If they offer smaller models for less money, it would eat up their premium offers.
The only way forward for American companies is to push for always bigger and more expensive models while using regulatory pressure and fear mongering (FUD) to discard price effective competition from China.
I don't known how Codex works, but we can set environment variables and point Claude CLI to deepseek. I think that before slashing prices they will slash those environment variables. After all they are not working to give a free TUI to deepseek and possibly to other competitors. But eventually yes, prices will go down or there will be an attempt at a regulatory capture.
Claude Code TUI is garbage. There's nothing worth protecting in there.
Honest question: what's garbage in that tool? I like that I can pair it with whatever editor I am using even if for 99% of the time I run it from inside emacs. I usually chat in the Claude window and it edits my files. I can also write a comment in the code, press a key and emacs will tell Claude to implement that comment. Simple needs, simple tool.
Most importantly, we need a model that doesn't randomly refuse us when we ask it to do something, or worse, deliberately sabotages us when it thinks we're building competing products. Like Anthropic's Fable.
I'm a fool for thinking that MiMo, in the context of Xiaomi who makes WiFi equipment (smartphones and routers), would be about network technology to manage parallel data streams (multiple inputs, multiple outputs https://en.wikipedia.org/wiki/MIMO)
I've worked a lot with MiMo in my project that pits LLMs against each other in games (clankerfights.ai). It is a very very good model for the price. MiniMax I'd say is smarter, but MiMo really touches near pareto frontier.
This is my favorite of the Chinese models I have tried. I think it would be hard to know if I was using Opus of MiMo if blindfolded in many instances.
Yes, but this has nothing to do with MiMo (the model).
This is what Claude Code is to Claude
I found it relevant and actually just the information I was looking for. Having a highly recommended model behind the tool makes it worth further investigation.
Yes, MiMo is a great model - but it works well with Claude Code - so I don't think the harness is really going to make it shine even more
Uh, what model do you think this is using?
MiMo Code is not a model, it's a harness like Claude Code / OpenCode / Codex (which is still open source, Apache 2.0, btw).
You might mean the MiMo-V2.5-Pro model?
He didn't say MiMo Code
Saying "This" on a post about "Mimo Code" leads one to, rightly, assume you are referring to "Mimo Code". If they commentor wanted to make a comment about a specific model it would have be clearer to simply mention the model instead of getting there transitively.
Mind numbing
Sorry for confusion. I indeed meant the model itself.
Good timing, I was looking for alternatives earlier today. opencode didn't install properly and I wasn't a fan of oh-my-pi and nanocoder.
MiMo code (via my z.ai coding plan) is very pleasant so far, nice UI and seems to respond faster than Claude Code. It might be injecting much less cruft into the conversation.
I also got access to the mimo-2.5-pro ultraspeed model yesterday, which is really quite snappy. It does cost more than DeepSeek, though, so I'm not sure whether it's worth it yet. Definitely fast though.
Opencode didn't install properly? Its just "mise use -g opencode@latest"
It did install, but then would hang when trying to configure the provider (specifically I was trying diffusiongemma via Nvidia). I gave up after a couple of attempts and moved to the next one.
is it local compatible and does it have telemetry?
it does have telemetry, enabled by default, that sends metrics to tracking.miui.com, including what model you are using. it can be turned off by environment variable (MIMOCODE_ENABLE_ANALYSIS=false), and yes it still has all the normal OpenCode provider logic so it will work with other/local models. it also automatically looks for updates and fetches a mimo model list, including when the telemetry is off, though those can also be disabled.
telemetry enabled by default and named "analysis" is not great.
"MiMoCode is built as a fork of OpenCode."
Why not just contribute to OpenCode instead of creating a clone :/
Because they want to optimize it for their models and don't want to be blocked by waiting for PRs to merge or be rejected.
There's plenty of reasons to start your own fork that you have full agency of, as long as the OSS License is maintained anyone will be able to benefit from any new features they want to make use of.
This is the beauty of open source :) KHTML -> WebKit -> Blink is a good example.
KHTML is dead now, though. It was basically embraced, extended and extinguished by Apple and Google, who both wanted to take away the leverage of the community.
Today, legacy KHTML maintainers are boxed-out of upstream decisions that might prevent Manifest v2 from swirling down the drain. I'd argue the story isn't very beautiful anymore.
True. Although the cruel twist is now that KDE's upstream decisions are boxing out X11/sysvinit/etc maintainers.
There are no X11 maintainers, there’s nobody to block out. In terms of users, of course there are X11 users. But nobody is maintaining it in any meaningful way and it’s been that way for years.
Also sysvinit is straight up obsolete software. Even if you hate systemd, there’s just no reason to use sysvinit because things like OpenRC and runit exist.
In addition, the adoption of systemd has lead to needing to maintain less init scripts because systemd units are just more portable. It used to be that every single distro and even versions of that distro required specialized init scripts for every application.
X11 and sysvinit are not downstream KParts libs like KHTML was. KDE has no obligation to fork or support either project.
KDE is not a fork, nevertheless KDE was building on X11 heavily and now they are trying to kill it.
So while technically not the same story as if it was a direct fork of X11, practically they have the 'obligation' to support X11 indefinitely. KDE trying to kill X11 too is the same exact EEE that you mentioned as a sad story in regads of KHTML.
Opencode sits on a ton of important PR's, so they didn't want to wait. Everybody else switched to omp (oh my pi) already.
To go a different path perhaps? You can't expect that all your ideas will land into a main repo and you really want to implement your vision while using a sane base.
OpenCode can merge in all their changes if they want.
There's a blog link https://mimo.xiaomi.com/blog/mimo-code-long-horizon
I think there's simply too much changed.
Because its currently impossible to "contribute to Opencode".
There are over 500 pages of open issues, up from 78 less than a month ago. They are doing nothing to halt the garbage/duplicates that pop up, and not even addressing legitimate PRs/reports.
Could just be a courtesy - Americans tend to be rather suspicious and hostile to contributions coming from China, and it might draw unwarranted attention from agencies and bad media.
Why not?
> Why not just contribute to OpenCode instead of creating a clone :/
It's controlled by a different organization; in particular a startup in a "competing" space.
have you ever tried contributing a large number of changes to OSS?
Unlimited context catches m :D I understand that it's not possible but it's worth to be checked to understand how you do your context compressions, etc.
I thought this was a wireless/MIMO radio project at first
yeah, was also expecting some disruption in the RF-design space.
Kinda RF-nerd clickbait... :)
Well Xiaomi is first and foremost a mobile phone company.
I also thought the same lol. It also happened with lora
I'm pretty sure most of the comments in this thread are AI bots propping up this product?
I tried the free model and it's nowhere near Sonnet 4.6 in terms of capabilities. The fact that token speed will randomly get stuck at 0/s makes sense given it's a free service, but the way it performs is more reminiscent of AI from 2025.
It's "just" an opencode fork but it adds some nice features to try out while not being a full orchestrator metapackage like oh-my-opencode. Quite nice! Though it would be even nicer if this stuff came upstream or as an easy extension instead in the future
> MiMo Code is a terminal-based coding agent built by Xiaomi's MiMo team on top of OpenCode and open-sourced under the MIT license.
I think it is great that they built it on top of open code. Open Code harness is good and I want it to grow. Harness is very important and more projects use it, the more it is adopted.
My coding agent VT Code has recently become a Xiaomi Orbit partner. If you want to try out Xiaomi Mimo V2.5 and V2.5 Pro in a different harness, feel free to use my VT Code. VT Code supports Mimo V2.5/Pro via official Xiaomi endpoints and via OpenRouter. Thank you!
[0] https://github.com/vinhnx/VTCode/blob/main/README.md#Provide...
> Unlimited Context
>Knowledge accumulates automatically with lossless compression, preserving every critical detail even across million-line projects.
As much as I absolutely love Mimo V2.5 Pro (it's a genuinely good model), I absolutely hate the way they calculate usage in their token plan.
For example: For a super small task in a small project that should not be consuming more than 500K total tokens after all tool calls included, their shown usage shot up to 152 million tokens.
But, when I scroll down on the same page, a table shows usage as 3 million tokens, out of which 2.5 million were cached.
This is such a huge conflict on the very same page. The bad thing is that the usage progress bar is shown against that 150 million token usage, not against that 3 million one.
This has been in discussions for at least past 3 months on reddit as well, and was precisely the reason I subscribed to their lowest tier, and for a single month only.
Update: their own harness, mimocode, shows total token usage as just 63.1K. We now have 3 entirely different values, differing in 3 orders of magnitude.
Update 2: So, I did the exact same task this time using DS4Pro, and total token usage was just 101K (as shown by opencode).
It's very confusing. They have tokens for their API and credits for their "token plan".
Even worse... they use both terms on the same page in dashboard.
"""
Credits 4,100,000,000 Credits
Total Token Consumption
"""
Ooooo... I now realize the trick. It's a mental play... give 1000x of "credits" but charge in same old tokens.
I prefer this to Anthropics "in the next 5 hours you have X tokens to spend, and we are not going to tell you what X is"
Here token price is model price + busy hours surcharge, and BTW nowhere near 1000x
No doubt this is better than Anthropics... and yes, I computed the ratio again: it's 38.92 (at least based on what's shown in my dashboard).
Yeah, by the way this is also opensource (do not run)
No 'training data', not 'open source'.
MiMo Code is a coding harness, not the model.
Only tangentially related: MiMo-2.5Pro is fast, cheap and very capable, although not quite gpt5.5 level iontelligence (I dont use the claudes). It works flawlessly in Pi and is an excellent workhorse. I expect big things from their next model.
I'm kind of surprised the demo UI is macOS. Are they mainly using Apple products to develop these things?
The more advanced devs all use apple laptops, sure.
All the Ai related companies use Macs.
Who isn’t?
I'm slapping debian on any crap hardware around, but that's just me with different ideological standards.
Even better to slap Debian on a 2nd hand ThinkPad :)
Also, Mario Zechner wrote libgdx and his first book on a netbook IIRC.
It's free to use because you are the product they are selling.
The installation method they officially propagate is dangerous. curl -fsSL https://mimo.xiaomi.com/install | bash
This is usually a PoC (Proof of concept) way to install something on a temporary container or temporary VM, but not for production use during daily desktop operation.
I was hoping their documentation would provide better installation instructions. But strangely, only for Windows do they recommend "npm install -g @mimo-ai/cli," which is a much better approach to managing installed packages.
For Mac/Linux, they have the strange recommendation to use the dangerous "curl <some_url> | bash." Quote:
> (for the best experience, Mac users are strongly encouraged to use iTerm or the VSCode Terminal) > curl -fsSL https://mimo.xiaomi.com/install | bash
:(
This is how everyone does it now. Including Anthropic.
To be fair, is that any different from naively trusting NPM? It's not like NPM is doing any vetting. They're every threat actors favorite sandbox these days.
https://code.claude.com/docs/en/quickstart
You're right that it's as dangerous as it's executing random third-party code on your machine, but the method also has propagated far beyond PoCs and such at this point. All of these projects and many others push that install method: Bun, Deno, rustup, k3s, Docker (if using their helper script), Homebrew, Tailscale...
Frankly, it's not really more insecure than any other installation method. Apt packages and the like generally have the ability to specify pre/post-install scripts, so `sudo dpkg -i ./random.deb` is equivalent to `sudo bash ./random.sh`. Even if they didn't have pre/post-install scripts, they're still writing arbitrary files to arbitrary locations on your disk, so they can trigger execution the next time you boot or log in or whatever.
And at the end of the day, no matter the installation method (even just unpacking a tarball and executing the program directly from that directory), you're going to run their program on your computer, and then the program can do whatever it wants. Maybe you don't run it with sudo, but https://xkcd.com/1200/ seems relevant.
A package (like a .deb) is a static artifact. It can be hashed, mirrored, and GPG-signed. Package managers usually verify that signature before any pre/post-install scripts. A "curl <some_url> | bash" pipe is a dynamic stream; the server can perform targeted attacks: sending a clean script to 99% of users and a malicious payload only to a specific IP address or User-Agent. This allows for targeted attacks that are invisible to the rest of the community.
Yes, running third-party code is always a leap of faith, but why choose a delivery method that removes the possibility of verification and opens the door to targeted injections? Convenience shouldn't be an excuse to ignore basic security hygiene.
The problem is that npm, cargo, etc. set the standard in people's minds for how package managers work, when the Linux community has been working on securing the supply chain issues for decades.
Like requiring a WoT (usually with physical meetups) vetting people creating packages, FTP-masters, dedicated clean buildbots, etc. in addition to the packages themselves being signed and so on.
Codex use this (for update).
> sh -c 'curl -fsSL https://chatgpt.com/codex/install.sh | CODEX_NON_INTERACTIVE=1 sh'
This is just sh, not bash, but I doubt it would be any better.
Thats exactly same as Claude Code offer: https://code.claude.com/docs/en/quickstart
We've had this discussion since Eazel Linux desktop popularized bash | curl in 2001.
> npm install ... is a much better approach to managing installed packages.
No. Until the upcoming version of npm is out, npm will also run arbitrary code. Almost all common installation tools run arbitrary code. Not doing that is sadly the exception for now.
Isn't executing arbitrary code kind of the entire point of NPM though? Any chance you have a link to something that describes their plans?
> Isn't executing arbitrary code kind of the entire point of NPM though?
No. npm is a package manager. As mentioned in the comment you're replying to, almost all package managers execute arbitrary code. Eg:
- pip
- Cargo
- apt/dpkg
- dnf/yum
- Homebrew
- RubyGems
- Composer (limited)
- Maven
> Any chance you have a link to something that describes their plans?
https://github.blog/changelog/2026-06-09-upcoming-breaking-c...
I get what you mean, but an NPM package is just a tarball of arbitrary code and some metadata. The whole point of it is to eventually run that arbitrary code, presumably. Otherwise why would you want to download the tarball and extract it? In fact, what purpose does NPM even serve if it's just a way to host tarballs?
I get the install time and run time execution might feel different, but I don't see how that's a security boundary at all.
I suspect that everyone will just get into the habit of typing --allowScripts all or whatever and nothing will actually change, because there's no point in a version of NPM that doesn't properly set things up for most people.
The code in the module isn’t arbitrary: it’s what the user intended an install, provided the package hasn’t been compromised. I do see your own version though.
Most apps don’t need install scripts so disabling them by default is fine.
I use npm 11.16.0 and it did this
npm warn allow-scripts Run `npm approve-scripts --allow-scripts-pending` to review, or `npm approve-scripts <pkg>` to allow.
Isn't Unlimited Context pretty difficult to promise? What exactly do they mean, could I just have two agents locked into a TTRPG back and forth forever?
Do you plan to ask them some master plan to live forever?
You can run the two agents in GAN like loop.. each trying to better the other. Give them a common task like design a better alternative to transformer model that uses max O(nlogn) memory, and the result comes closest to existing n^n implementations.
Good idea actually.. why haven't I tried this before.
That is an incredibly annoying grunge font. And what is the point of the hidden image in the background that reveals under your mouse cursor.
It's interesting that it renders Chinese in a TUI. I wonder if that breaks anything that assumes a character is always a column wide.
Terminals and things that live in terminals have relied on wcwidth() to handle this since time immemorial (which is always fun when they are out of sync, e. g. remote over ssh, but in the vast majority of the cases it just works).
This is super exciting, can't wait to try it out
macOS binary (mimocode-darwin-arm64.zip ) seems broken: "“mimo” is damaged and can’t be opened. You should move it to the Trash."
No, you are just experiencing the best of Apple. How dare you download non notarized binaries on your own computer? Do you have a license for that?
Terminal > sudo xattr -rd com.apple.quarantine > Drag and drop the app into terminal > enter and enter your password
Strange, I reckon I installed Tahoe just a while ago and still didn't have a similar issue, but I remember on previous MacOS versions the error message for unotarized binaries, was to warn that there was indeed a security issue with the binary, not that it was simply "damaged".
A bit crappy on Apple's side.
Thank you.
"damaged" by not paying 99 bucks.
"damaged", even after paying 99$ : https://github.com/electron/notarize/issues/205#issuecomment...
See last 20'ish comments.
the OS is broken, in this case.
Are AI people using LLMs to name things? Just take a widely popular thing that already means something, capitalise it differently, and you're done.
Microsoft's LoRA (already a thing called LoRa) and now MiMo (already a thing called MIMO)
Maybe a classic Google search is not so bad, eh?
Only worked for about 5m, then Too many requests.
mimocode gets it. This is actually, impressive! Chinese models are really up there with the rest.
Reading "xiaomi"in the headline I was thrilled to see MiMo only to find out it's NOT this one: https://en.wikipedia.org/wiki/MIMO What a disappointment!
Looks an awful lot like OpenCode
> MiMoCode is built as a fork of OpenCode.
That’s why
Why is OpenCode awful?
He didn't say that.
https://www.merriam-webster.com/dictionary/an%20awful%20lot
Who said that?
looks great. surprised that Xiaomi has made such great advancements in AI
Redditors are unhappy about their coding plan: https://www.reddit.com/r/opencodeCLI/comments/1t37dz3/xiaomi...
I guess the way to use their models is through another provider, like https://opencode.ai/go
Things have improved a lot since last month. Now their cheapest plan gives you 4.1B credits and I think the cache situation has improved too.
My comment on same: https://news.ycombinator.com/item?id=48493358
Yes the subscription plan costs literally the same thing as just paying API-per-token pricing.
isn't this just opencode with a different logo?
I got an invite to test their ultra fast model only to be geofenced when trying to use it. Pff!
Yeah that sucks. Now imagine the average Chinese developer who encounters this all the time!
Let the battle for the harnesses give free tokens for all, until the next competition arena does the same. It seems that's the only way AI will remain acessible.
This website is gorgeous, by the way. The mouse reveal on the background, amazing.
Can this be used as an alternative to Claude backend? For Ralph loops? Replacing `claude -p`? Anyone can shed a light on this?
Hm, can I just use free tokens without using MiMo-Code?
OpenCode or pi.dev are enough. I don't like CC-style agent lock-in, regardless if it's Anthropic or Xiaomi doing it.
Any english links?
You can change language from the top right-most dropdown, and select English
Top right corner
I wonder what the minimum required memory specification is
Is that Open-Source like, run it locally, no phone home included, or open source like the thin front-end layer is all that is actually open-source but it’s an empty shell without the remote API it relies on?
They default it to talking to a free version of their model (which is incredibly cheap if you decide you like it.)
But it seems trivially easy to run it against local models. Their onboarding guide offers that option, though I have no idea if it changes any functionality.
It defaults to the API, but you can download the weights for the 1T-parameter Pro model and selfhost if you have the hardware for it.
The latter. It looks like it's meant to be a batteries-included agent to promote their free-for-a-limited time AI service that it connects to by default.
Ok, fair enough compared to the rest of the proeminent actors I guess, but quite confusing from dev point of view. Lately I started to experiment with model like Qwen2.5 on local. Good enough to ask simple question, but didn’t manage to do anything remotely close a agents I started to experiment with through Copilot.
qwen3.5 9b runs okay on my 12GB gaming GPU. It's very stupid as a coding agent but it's possible to get useful work out of it.
I am experimenting with LFM2.5-8B-1A and getting 250tps on a 3060
sounds really cool if it was coming out of anywhere except China, which has laws to exfiltrate your data and send it back to the government for espionage purposes [0].
[0] https://www.fbi.gov/investigate/counterintelligence/the-chin...
Today on open source vampires: xaomi forks an open source project, doesn't contribute upstream, attaches usage restrictions that are probably incompatible with the license, and wants good PR. Fuck these people.
Open source and open weight AI is very important to protect freedom of speech. OpenAI, and ESPECIALLY Anthropic, will try to ban them through regulatory capture / safety fearmongering. We need to make sure that does not happen. It’s not society’s problem if these frontier labs have no moats.
You know they are benchmaxxing when they end up writing their coding harness in TypeScript npm slop
Their models can't help them build it with something better?
That's the only benchmark people need, whether or not their model can raise the bar of their own product
And so far it's looking pretty sad
It was already open-source `https://github.com/anomalyco/opencode`