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.
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
> 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)?
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?
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.
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.
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.
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. 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, but 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.
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 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.
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.
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 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.
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.
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'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.
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.
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.
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.
I don't think that's true? AFAIK OpenCode started as a TUI and their GUI app is Tauri-based, so don't think it was forked from OpenCode. You might be thinking of Cursor
There were once two harnesses named OpenCode, one written in Go & the other in TypeScript (the more popular one).
Kujtim Hoxha creates a project named TermAI. (SST folks) Dax & Adam join the project, rebrand it to OpenCode with Dax buying the domain, opencode.ai.
Charm, the company behind the original libraries, acqui-hires Kujtim, who moves the project to Charm's organization, leaving SST unimpressed (due to VC involvement?)
Allegations Charm rewrote git history and deleted GitHub comments.
Dax claims ownership of the brand, forks project. For a time, 2 projects named OpenCode exist. Charm eventually renames its version to "Crush".
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?
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".
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 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.
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.
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.
From github
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)?
For example, I can send screenshots of what I'm developing and it understands.
That was a good call. Gemini CLI is dead.
“ 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
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
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?
How many sites do this but you don't notice because they default to English?
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.
Based solely on quality and price, OpenAI, Anthropic, and other western models just can't compete with the new generation of Chinese open models.
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, but 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.
This is what Claude Code is to Claude
You might mean the MiMo-V2.5-Pro 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
:(
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
> 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.
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.
> 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.
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...
Why not just contribute to OpenCode instead of creating a clone :/
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.
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.
I think there's simply too much changed.
It's controlled by a different organization; in particular a startup in a "competing" space.
OpenCode started as an independent CLI project. Their desktop app is still in beta, and it was never a fork of VS Code.
I believe they contain no code derived from VS Code.
Kinda RF-nerd clickbait... :)
>Knowledge accumulates automatically with lossless compression, preserving every critical detail even across million-line projects.
Terminal > sudo xattr -rd com.apple.quarantine > Drag and drop the app into terminal > enter and enter your password
A bit crappy on Apple's side.
Thank you.
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.
telemetry enabled by default and named "analysis" is not great.
OpenCode or pi.dev are enough. I don't like CC-style agent lock-in, regardless if it's Anthropic or Xiaomi doing it.
That’s why
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.