A tier list of China's top 19 open model builders. Who did we miss? At the frontier * DeepSeek * Qwen Close competitors * Moonshot AI (Kimi) * Zhipu / Z AI Noteworthy * StepFun * Tencent (Hunyuan) * RedNote (Xiaohongshu) * MiniMax * OpenGVLab / InternLM * Skywork On the rise * ByteDance Seed * OpenBMB * Xiaomi (MiMo) * Baidu (ERNIE) Honorable Mentions * Multimodal Art Projection * Alibaba International Digital Commerce Group * Beijing Academy of Artificial Intelligence (BAAI) * inclusionAI * Pangu (Huawei) I learned a lot from these. We have so much more we need to do to understand how their AI ecosystem works.
Interconnects
 @interconnectsai
Aug 17
China's Top 19 Open Model Labs We ranked all the organizations in China releasing open models, from the top of DeepSeek to small, newer academic labs making waves with tech reports and niche models. interconnects.ai/p/chinas-to…

Aug 17, 2025 · 3:49 PM UTC

46
170
24
1,073
This is based on open outputs, which is all we can measure, even though operationally I bet this list could shuffle a ton.
3
27
RE ByteDance Seed
1
21
I'd love to actually visit all the top labs and make a judgement on pure quality of internal and closed artifacts too, but that's not the life Im in
1
1
30
The strong opinions about this are leaking a lot of alpha previously restricted to private group chats, im enjoying this.
2
1
24
Replying to @natolambert
We will keep up our efforts!
1
1
40
Replying to @natolambert
which of the small labs were you most surprised by? e.g. openbmb has about ~100 staff w leaders from tsinghua like @zibuyu9 - they've done impressive stuff with making multimodal work on mobile devices
4
Replying to @natolambert
Interesting rankings. The open-source landscape is evolving so rapidly, even this list may soon change. Who will rise next?
1
Replying to @natolambert
Need this list for Chinese robotics
1
Replying to @natolambert @bgurley
Seed. And Tencent more that just noteworthy..
Replying to @natolambert
Qwen is the best overall, deepseek more researchy
14
Replying to @natolambert
Kimi K2 is S-Tier
10
Replying to @natolambert
It's outrageous that everyone keeps considering Deepseek frontier. They delivered twice that's it
3
1
4
4
Replying to @natolambert
I reluctantly accede that Moonshot's Kimi k2 is currently a step below in terms of technical competence and accuracy (resisting hallucination). It's still so delightful though. So diverse and interesting. If their k2-reasoner comes out great, you might have to upgrade them.
5
Replying to @natolambert
zai's models are just as good at coding as Opus/Sonnet but far more affordable in terms of pay-per-token but also have prompt caching. GLM 4.5 costs ~7x less than sonnet and is just as good as a developer. What I'm seeing these Chinese labs ship is value, while many American labs are shipping rhetoric. The recent open-source model from OpenAI basically maxed out the meh level and despite the safety rhetoric was actually easier to prompt inject than the smaller, and better, glm 4.5 air. Going forward, I will simply not financially support closed AI labs or even use models with weird licenses like gemma, llama and kimi. If I couldn't at least conceptually self host a model, or the model has weird bs terms in the license, then it is just completely off the table for me.
1
4
Replying to @natolambert
Where's the list of of American open models?
1
4
Replying to @natolambert
ByteDance is the top place grads choose, across the board. It’s far more dominant than I realized (and Tech Twitter seems to know). Rednote is closest to Reddit (not Instagram as I originally thought). It’s the go-to for community answers. Wouldn’t say they do frontier tech though.
1
3
Replying to @natolambert
Qwen is by alibaba. You also missed Tencent, they released the Hunyuan 3d models.
3
Replying to @natolambert
Z AI is at the frontier. Glm 4.5 is eye opening.
2
Replying to @natolambert
Kimi K2 isn't at the top because it's not a reasoning model? Bruh...
2
Replying to @natolambert
Top 3 1 - qwen 2 - Z ai 3 - deep seek
2
Replying to @natolambert
This needs the meme where if you pull the cover over the models, it’s all distilled western models behind it.
4
1
Replying to @natolambert
Wow they are popping up fast!
1
Replying to @natolambert
Bytedance Doubao
1
Replying to @natolambert
Minimax is S tier
1
Replying to @natolambert
Baidu is sorta pathetic considering it’s one of the earliest in the game. Amodei and Andrew Ng had worked in that lab.
1
Replying to @natolambert
Vivo also has a open weights model blueLM I have not tried it but you check in future!
1
Replying to @natolambert
Bytedance is more then on the rise…
1
Replying to @natolambert
Kimi 2 not frontier? Cmon dawg
1
Replying to @natolambert
Vivo is also doing great work in computer vision domain github.com/orgs/vivoCameraRe…

vivo Camera Research

vivo Camera Research has 15 repositories available. Follow their code on GitHub.

github.com
1
Replying to @natolambert
China's AI ecosystem grows relentlessly—DeepSeek and Qwen now rival global peers, while smaller labs like Skywork push niche innovation.
1
Replying to @natolambert
Isn't ByteDance Seed mostly based in the US ?
2
Replying to @natolambert
@vivago_ai with the hi dream models
Replying to @natolambert
Need to start testing more models, I haven't heard about a lot of them
Replying to @natolambert
Minimax is tier 0
Replying to @natolambert
I don't know whats good with deepseek,deepseek can't even understand a picture, z.ai for me far more ahead. Or someone can explain whats good on deepseek?
Replying to @natolambert
This is insightful, they're leading on the open source front
Load more