# ![@Xianbao_QIAN Avatar](https://lunarcrush.com/gi/w:26/cr:twitter::1597257798068637697.png) @Xianbao_QIAN Tiezhen WANG

Tiezhen WANG posts on X about model, ai, llm, the new the most. They currently have [-----] followers and [---] posts still getting attention that total [-----] engagements in the last [--] hours.

### Engagements: [-----] [#](/creator/twitter::1597257798068637697/interactions)
![Engagements Line Chart](https://lunarcrush.com/gi/w:600/cr:twitter::1597257798068637697/c:line/m:interactions.svg)

- [--] Week [-------] +632%
- [--] Month [-------] +7.30%
- [--] Months [---------] +218%
- [--] Year [---------] +246%

### Mentions: [--] [#](/creator/twitter::1597257798068637697/posts_active)
![Mentions Line Chart](https://lunarcrush.com/gi/w:600/cr:twitter::1597257798068637697/c:line/m:posts_active.svg)

- [--] Week [--] -7.70%
- [--] Month [--] +48%
- [--] Months [---] +36%
- [--] Year [---] +84%

### Followers: [-----] [#](/creator/twitter::1597257798068637697/followers)
![Followers Line Chart](https://lunarcrush.com/gi/w:600/cr:twitter::1597257798068637697/c:line/m:followers.svg)

- [--] Week [-----] +2.90%
- [--] Month [-----] +5%
- [--] Months [-----] +47%
- [--] Year [-----] +118%

### CreatorRank: [-------] [#](/creator/twitter::1597257798068637697/influencer_rank)
![CreatorRank Line Chart](https://lunarcrush.com/gi/w:600/cr:twitter::1597257798068637697/c:line/m:influencer_rank.svg)

### Social Influence

**Social category influence**
[technology brands](/list/technology-brands)  17.24% [finance](/list/finance)  6.03% [social networks](/list/social-networks)  2.59% [stocks](/list/stocks)  2.59% [celebrities](/list/celebrities)  1.72% [countries](/list/countries)  1.72% [musicians](/list/musicians)  0.86% [exchanges](/list/exchanges)  0.86% [automotive brands](/list/automotive-brands)  0.86%

**Social topic influence**
[model](/topic/model) #3090, [ai](/topic/ai) 11.21%, [llm](/topic/llm) #116, [the new](/topic/the-new) 6.9%, [release](/topic/release) 6.9%, [up to](/topic/up-to) 6.9%, [strong](/topic/strong) #1681, [performance](/topic/performance) 5.17%, [link](/topic/link) 5.17%, [super](/topic/super) #3773

**Top accounts mentioned or mentioned by**
[@huggingface](/creator/undefined) [@deepseekai](/creator/undefined) [@zaiorg](/creator/undefined) [@tencenthunyuan](/creator/undefined) [@alibabaqwen](/creator/undefined) [@txhunyuan](/creator/undefined) [@stepfunai](/creator/undefined) [@bytedancetalk](/creator/undefined) [@meituanlongcat](/creator/undefined) [@sunnypause](/creator/undefined) [@kimimoonshot](/creator/undefined) [@baidu](/creator/undefined) [@thukeg](/creator/undefined) [@chatglm](/creator/undefined) [@alibabagroup](/creator/undefined) [@openmmlab](/creator/undefined) [@tongyispeechai](/creator/undefined) [@hailuoai](/creator/undefined) [@internlm](/creator/undefined) [@xpengmotors](/creator/undefined)
### Top Social Posts
Top posts by engagements in the last [--] hours

"SeedDance [---] looks super good. Apart from the great motion and details it's also acoustic Looking forward to the model. @BytedanceTalk Is there a way to apply for beta testing"  
[X Link](https://x.com/Xianbao_QIAN/status/2019931264913117388)  2026-02-07T00:28Z [----] followers, 243.9K engagements


"Oh What's next"  
[X Link](https://x.com/Xianbao_QIAN/status/2021144428317221296)  2026-02-10T08:49Z [----] followers, 121.8K engagements


"Oh I need this model in my OpenClaw SoulX-Singer a great zero-shot singing voice synthesis model has been released on @huggingface Apache [--] license as well Check out the demo: And the model: https://huggingface.co/Soul-AILab/SoulX-Singer https://huggingface.co/spaces/multimodalart/SoulX-Singer https://huggingface.co/Soul-AILab/SoulX-Singer https://huggingface.co/spaces/multimodalart/SoulX-Singer"  
[X Link](https://x.com/Xianbao_QIAN/status/2021475556437655659)  2026-02-11T06:45Z [----] followers, 21K engagements


"ByteDance is cooking The two LLMs by their Seed team both look super strong. Seed [---] Pro released. Reaching the Frontier. Model Card: https://t.co/zmyZWrYlwv https://t.co/uBgGKFXd66 Seed [---] Pro released. Reaching the Frontier. Model Card: https://t.co/zmyZWrYlwv https://t.co/uBgGKFXd66"  
[X Link](https://x.com/Xianbao_QIAN/status/2022623027100622891)  2026-02-14T10:44Z [----] followers, 16.1K engagements


"Please tell me this isn't AI generated. PS: I was so convinced it was that I actually went looking for proof of fraud. Turns out it's made by a human director. We've reached the point where real art looks suspicious. Ai https://t.co/9Rdmyj5R9o Ai https://t.co/9Rdmyj5R9o"  
[X Link](https://x.com/Xianbao_QIAN/status/2022625827813495214)  2026-02-14T10:55Z [----] followers, [----] engagements


"PS: Having a wechat account helps you work with the open source models better. For instance this great model from meituan only mentioned WeChat group for technical support :) New model from the @Meituan_LongCat family LongCat-Flash-Thinking-2601 - 560A27B MoE model - MiT license weights - Demo available on with thinking mode - Basic @sgl_project & @vllm_project support merged https://longcat.ai https://longcat.ai"  
[X Link](https://x.com/Xianbao_QIAN/status/2011530069295186252)  2026-01-14T20:05Z [----] followers, [----] engagements


"Ive been seeing a growing number of arch innovations coming out of CN models from @Alibaba_Qwen @deepseek_ai @Zai_org @Meituan_LongCat @RWKV_AI and many others The spirit of open-sourcing is to enable exploration from every angle; looking forward to your next releases this year i would like to slow things down a little bit and make it better. invest more on research that might take u to nothing this year i would like to slow things down a little bit and make it better. invest more on research that might take u to nothing"  
[X Link](https://x.com/Xianbao_QIAN/status/2012685428818903175)  2026-01-18T00:36Z [----] followers, [----] engagements


"Erine [---] Technical report is out https://arxiv.org/pdf/2602.04705 https://arxiv.org/pdf/2602.04705"  
[X Link](https://x.com/Xianbao_QIAN/status/2019322982917943429)  2026-02-05T08:11Z [----] followers, [----] engagements


"@sunnypause I doubt about that"  
[X Link](https://x.com/Xianbao_QIAN/status/2020096917217427812)  2026-02-07T11:26Z [----] followers, [----] engagements


"OMY Is this AI generated seedance [---] ai Seedance 2.0AI https://t.co/BOumeGnDLv seedance [---] ai Seedance 2.0AI https://t.co/BOumeGnDLv"  
[X Link](https://x.com/Xianbao_QIAN/status/2020103169557295215)  2026-02-07T11:51Z [----] followers, 15.1K engagements


"Claude [---] seems to be really good at docx writing even when I don't need it 🤣 Me: write a report Claude: OKAY creating a comprehensive document in docx. [--] minutes later context length exploded"  
[X Link](https://x.com/Xianbao_QIAN/status/2020625584746094721)  2026-02-08T22:27Z [----] followers, [----] engagements


"New open source robotic arm from @seeedstudio with CNC parts and @RobStride_com motors https://www.bilibili.com/video/BV1DJ6KBNEgD/vd_source=16f6410bb6a7c9939bfb6a93fc79b27e https://github.com/Seeed-Projects/reBot-DevArm https://www.bilibili.com/video/BV1DJ6KBNEgD/vd_source=16f6410bb6a7c9939bfb6a93fc79b27e https://github.com/Seeed-Projects/reBot-DevArm"  
[X Link](https://x.com/Xianbao_QIAN/status/2020652201077805115)  2026-02-09T00:13Z [----] followers, [----] engagements


"Kimi is killing the game and becoming the most used @openclaw model based on @OpenRouterAI metrics. Great work @Kimi_Moonshot @opencode @fanjiewang"  
[X Link](https://x.com/Xianbao_QIAN/status/2020680686727004180)  2026-02-09T02:06Z [----] followers, 11.9K engagements


"DeepSeek has reportedly released a new model on @deepseek_ai mobile app once upgraded to version 1.7.4(1). The new model is said to be significantly better and has 1M context length. Report: PS: I'm still on 1.7.3(169) not sure how to upgrade though. https://mp.weixin.qq.com/s/0Eio50b2JNGQV-gUBYbd3Q https://mp.weixin.qq.com/s/0Eio50b2JNGQV-gUBYbd3Q"  
[X Link](https://x.com/Xianbao_QIAN/status/2021519824845963450)  2026-02-11T09:40Z [----] followers, 19.5K engagements


"Wait. is the AI Video race actually over Not an expert on video generation but just thinking out loud if this level music quality is generated directly out of the video (which is case based on close lip sync IMO) it feels that the loop has closed. Weve moved past "can it be done" to "how cheap and easy can we make it." Aside from more industry problems including longer frames better text faster generation whats the next actual frontier Or is this it wdyt Seedance made this MV ft. Kayne West. Its so well made that its going viral on Chinese internet. https://t.co/vUrlgIqVeX Seedance made this"  
[X Link](https://x.com/Xianbao_QIAN/status/2021699691701055771)  2026-02-11T21:35Z [----] followers, [----] engagements


"Great movie generated by AI with real human voice (model unknown but not Seedream2) It should be made to a game Originally posted here: https://www.bilibili.com/video/BV1hvH6zTEhH/vd_source=c5a5204620e35330e6145843f4df6ea4 Found this Mulan stylized short made with Seedance [---] https://t.co/PkyGWthzKk https://www.bilibili.com/video/BV1hvH6zTEhH/vd_source=c5a5204620e35330e6145843f4df6ea4 Found this Mulan stylized short made with Seedance [---] https://t.co/PkyGWthzKk"  
[X Link](https://x.com/Xianbao_QIAN/status/2021702511569383912)  2026-02-11T21:46Z [----] followers, [----] engagements


"@baidu is back Two models and a new global web portal has been released today. Finally the phone verification is no longer needed They do listen to the voice of developers. This model release includes both enhanced lightweighted thinking model ERNIE-4.5-21B-A3B-Thinking already available on @huggingface now. It also includes a new stealth Ernie X [---] model that is hidden in the UI and claims to smarter and think deeper. I tried generating a simple Snake game. When did coding a Snake game become the ultimate vibe-coding [---] test The result actually looks great. Im really into the clean design"  
[X Link](https://x.com/Xianbao_QIAN/status/1965450400531964159)  2025-09-09T16:21Z [----] followers, 13.9K engagements


"A new "stealth" model dropped on OpenRouter last nightPony Alpha. Free 200K context [--] TPS and suspiciously good at coding. But who actually built it Here's some speculations. 👇 200K context window rules out Gemini which has typically very long context length [--] TPS throughput is too slow for frontier Western labs' API limits OpenAI Anthropic and Moonshot (Kimi) all just shipped major releases so they're unlikely to be the candidate What do you think https://twitter.com/i/web/status/2020095080011559270 https://twitter.com/i/web/status/2020095080011559270"  
[X Link](https://x.com/Xianbao_QIAN/status/2020095080011559270)  2026-02-07T11:19Z [----] followers, 10.9K engagements


"@real_kai42 K2.5 GLM5 MiniMax 2.5"  
[X Link](https://x.com/Xianbao_QIAN/status/2022299556869030215)  2026-02-13T13:19Z [----] followers, [---] engagements


"Open weight LLM is not the same as OS software since theres no developer community contributing back. It does exist but in different form. Derived models enhancing the original one on certain areas is the Software [---] style "contributing back" let alone knowledge sharing or optimizations (e.g. quantization / infra work). Anthropic CEO Dario - - Deepseek - https://t.co/LscojWCJCf Anthropic CEO Dario - - Deepseek - https://t.co/LscojWCJCf"  
[X Link](https://x.com/Xianbao_QIAN/status/1970430148253229355)  2025-09-23T10:08Z [----] followers, 18.5K engagements


"MiniCPM o [---] a 9B model built on top of SigLip2 Whisper-medium CosyVoice2 and Qwen3-8B with the capability to do full-duplex communication directly on Mac. Link: https://huggingface.co/openbmb/MiniCPM-o-4_5 https://huggingface.co/openbmb/MiniCPM-o-4_5"  
[X Link](https://x.com/Xianbao_QIAN/status/2019762464074952944)  2026-02-06T13:17Z [----] followers, [----] engagements


"Options available: - ByteDance @ByteDanceOSS - Tencent @TencentHunyuan - @Zai_org - DeepSeek @deepseek_ai http://Z.ai http://Z.ai"  
[X Link](https://x.com/Xianbao_QIAN/status/2020097127482028346)  2026-02-07T11:27Z [----] followers, [----] engagements


"Pony Alpha is open sourced finally. Welcome GLM [--] by @Zai_org on @huggingface - 744B parameters (40B active) - 28.5T pretrain tokens - DeepSeek Sparse Attention (DSA) - Trained by asynchronous RL infrastructure slime - MIT license - Native FP8 version available https://huggingface.co/zai-org/GLM-5 https://huggingface.co/zai-org/GLM-5"  
[X Link](https://x.com/Xianbao_QIAN/status/2021682818636820781)  2026-02-11T20:28Z [----] followers, 21.2K engagements


"FARS (Fully Automated Research System) is an end-to-end AI research system. It autonomously conducts literature review generates hypotheses writes code runs experiments and produces research papers without human intervention. Watch out for the live research at tmr https://analemma.ai/fars Today were introducing FARS a Fully Automated Research System. Tomorrow at 10:00 PM Eastern Time well begin its first public deployment as a live experiment. During the deployment FARS will run continuously and autonomously aiming to produce [---] complete research https://t.co/kjmzj1MsKO"  
[X Link](https://x.com/Xianbao_QIAN/status/2021915100391870922)  2026-02-12T11:51Z [----] followers, [----] engagements


"The AI researcher has woken up https://analemma.ai/fars FARS (Fully Automated Research System) is an end-to-end AI research system. It autonomously conducts literature review generates hypotheses writes code runs experiments and produces research papers without human intervention. Watch out for the live research at https://analemma.ai/fars FARS (Fully Automated Research System) is an end-to-end AI research system. It autonomously conducts literature review generates hypotheses writes code runs experiments and produces research papers without human intervention. Watch out for the live research"  
[X Link](https://x.com/Xianbao_QIAN/status/2022153641684349245)  2026-02-13T03:39Z [----] followers, [----] engagements


"AIDC-AI/Ovis2.6-30B-A3B is out on @huggingface https://huggingface.co/AIDC-AI/Ovis2.6-30B-A3B https://huggingface.co/AIDC-AI/Ovis2.6-30B-A3B"  
[X Link](https://x.com/Xianbao_QIAN/status/2022182401812967474)  2026-02-13T05:33Z [----] followers, [----] engagements


"ByteDance just dropped Seedance [--] Seedream [--] and Seed [--] LLM before Chinese New Year like it was nothing. Everyone's talking about the DeepSeek moment. I'd argue the real moment this year belongs to @ByteDance who quietly is the most underrated AI lab on the planet. Now imagine if they went a little bit more open source. Happy to host them on @huggingface ByteDance is cooking The two LLMs by their Seed team both look super strong. https://t.co/fCs2UtpRWy ByteDance is cooking The two LLMs by their Seed team both look super strong. https://t.co/fCs2UtpRWy"  
[X Link](https://x.com/Xianbao_QIAN/status/2022633370841420180)  2026-02-14T11:25Z [----] followers, 12.3K engagements


"qwen2 is here. Apache [--] licensed model that beats LLAMA3 in bunch of bunch marks. It also talks in [--] languages. It comes with different sizes from 0.5B all the way to 72B give it a try on @huggingface : Demo: https://huggingface.co/spaces/Qwen/Qwen2-72B-Instruct https://huggingface.co/collections/Qwen/qwen2-6659360b33528ced941e557f https://huggingface.co/spaces/Qwen/Qwen2-72B-Instruct https://huggingface.co/collections/Qwen/qwen2-6659360b33528ced941e557f"  
[X Link](https://x.com/anyuser/status/1798751321518141487)  2024-06-06T16:18Z [----] followers, 14K engagements


"Been waiting for this A great model a day keeps GPT away. Here introduce DeepSeek-Coder-V2 236B MoE (21B active) [---] programming languages 128k context length. performance between GPT4o and GPT-4-Turbo on coding and math. https://github.com/deepseek-ai/DeepSeek-Coder-V2/blob/main/paper.pdf https://huggingface.co/deepseek-ai/DeepSeek-Coder-V2-Instruct https://github.com/deepseek-ai/DeepSeek-Coder-V2/blob/main/paper.pdf https://huggingface.co/deepseek-ai/DeepSeek-Coder-V2-Instruct"  
[X Link](https://x.com/anyuser/status/1802683736737427727)  2024-06-17T12:44Z [----] followers, 16.6K engagements


"What a week for video generation After the hype around closed-source models from Kling Luma and Runway Here comes the open-source one: Open-Sora [---] from @HPCAITech just got released @huggingface Highlights: 1.1B model trained on 20M data generate up to 14s 720p video"  
[X Link](https://x.com/anyuser/status/1802974444614476062)  2024-06-18T07:59Z [----] followers, [----] engagements


"Fish Audio v1.2 just got released on @huggingface Very impressive TTS results on Chinese https://huggingface.co/fishaudio/fish-speech-1.2 https://huggingface.co/fishaudio/fish-speech-1.2"  
[X Link](https://x.com/Xianbao_QIAN/status/1808127903810703839)  2024-07-02T13:17Z [----] followers, 82.4K engagements


"Would this be one of the best local LLM under 10B Or best small server side model with up to 1M context length InternLM-2.5 7B family just dropped on @huggingface License: Free commercial usage upon request with blanket approval. https://huggingface.co/collections/internlm/internlm25-66853f32717072d17581bc13 https://huggingface.co/collections/internlm/internlm25-66853f32717072d17581bc13"  
[X Link](https://x.com/Xianbao_QIAN/status/1808476861141545292)  2024-07-03T12:24Z [----] followers, [----] engagements


"The compeitition in the video portait field is so intense Here comes a new great model from Alipay team under AntFinance. Key words: Pose + Audio driven Talking/Sing Chinese + English Homepage: Model on @huggingface: https://huggingface.co/BadToBest/EchoMimic https://badtobest.github.io/echomimic.html https://huggingface.co/BadToBest/EchoMimic https://badtobest.github.io/echomimic.html"  
[X Link](https://x.com/Xianbao_QIAN/status/1810980157504766213)  2024-07-10T10:11Z [----] followers, [----] engagements


"EchoMimic is now available on @huggingface with A100 thx to @zhouwenmeng 's great work Give it a try on https://huggingface.co/spaces/BadToBest/EchoMimic https://huggingface.co/spaces/BadToBest/EchoMimic"  
[X Link](https://x.com/anyuser/status/1819156227966095844)  2024-08-01T23:40Z [----] followers, 12.7K engagements


"The new virtual try-on model from Kuaishou kcolors is very impressive. https://huggingface.co/spaces/Kwai-Kolors/Kolors-Virtual-Try-On https://huggingface.co/spaces/Kwai-Kolors/Kolors-Virtual-Try-On"  
[X Link](https://x.com/anyuser/status/1828064701810717018)  2024-08-26T13:39Z [----] followers, 21.7K engagements


"The best open-access video generation model CogVideoX-5B has been released by @thukeg They also had a 2B model in Apache [---] Amazing Give it a try on @huggingface Model: Demo: https://huggingface.co/spaces/THUDM/CogVideoX-5B-Space https://huggingface.co/THUDM/CogVideoX-5b https://huggingface.co/spaces/THUDM/CogVideoX-5B-Space https://huggingface.co/THUDM/CogVideoX-5b"  
[X Link](https://x.com/anyuser/status/1828402971622940781)  2024-08-27T12:03Z [----] followers, [----] engagements


"Open Sora Plan has released the [---] version of their video generation model. Love this #BlackMythWukong story illustrated by AI https://github.com/PKU-YuanGroup/Open-Sora-Plan/blob/main/docs/Report-v1.3.0.md https://github.com/PKU-YuanGroup/Open-Sora-Plan/blob/main/docs/Report-v1.3.0.md"  
[X Link](https://x.com/anyuser/status/1846554046729232585)  2024-10-16T14:09Z [----] followers, 33K engagements


"A new video generation model coming from @rhymes_ai_ Model Blog - Apache [---] - 175M parameter VideoVAE and a 2.8B parameter VideoDiT model - 6s videos at [--] FPS with 720x1280 resolution with up coming @diffuserslib integration https://huggingface.co/blog/RhymesAI/allegro https://huggingface.co/blog/RhymesAI/allegro"  
[X Link](https://x.com/anyuser/status/1848540874441150791)  2024-10-22T01:44Z [----] followers, 15.6K engagements


"End-to-End Speech / text model GLM-4-Voice from @ChatGLM - Support both Chinese and English - Tokenizer fine-tuned from Whisper encoder - Decoder based on CosyVoice to convert discrete tokens to speech Homepage @huggingface : https://huggingface.co/collections/xianbao/glm-4-voice-671b337123be0786ec31d231 https://github.com/THUDM/GLM-4-Voicetab=readme-ov-file https://huggingface.co/collections/xianbao/glm-4-voice-671b337123be0786ec31d231 https://github.com/THUDM/GLM-4-Voicetab=readme-ov-file"  
[X Link](https://x.com/Xianbao_QIAN/status/1849692235182608860)  2024-10-25T05:59Z [----] followers, [----] engagements


"@AlibabaGroup just unveiled the hidden secret of ChatGPT o1 model. They just released a o1 alternative with Apache [--] license. arco-o1 is powered by CoT fine-tuning MCTS reflection and reasoning optimized for complex real-world problem-solving tasks"  
[X Link](https://x.com/anyuser/status/1859808310700146795)  2024-11-22T03:57Z [----] followers, 15.1K engagements


"The Chinese community is roaring. Nobody is sure how o1 is made in OpenAI and after @deepseek_ai and @AlibabaGroup here comes the third approach from @OpenMMLab Key takeaways: - small model distilled from large model CoT results - a combination of general and expert models"  
[X Link](https://x.com/Xianbao_QIAN/status/1861023307799339115)  2024-11-25T12:25Z [----] followers, 31.3K engagements


"SUPER impressive FILM made by @TXhunyuan AI video generation model made by z. Where is sora Do you think that it will do much better"  
[X Link](https://x.com/Xianbao_QIAN/status/1863482475307249794)  2024-12-02T07:16Z [----] followers, 29.7K engagements


"The model is now available on @huggingface Key features: - 13B parameters the largest OS video generation model - Outperform Runway Kling Luma and Haikuo - Prompt rewrite model provided - at least 60GB vram for 720px1280px129f - Is this the end for game card players 😂 SUPER impressive FILM made by @TXhunyuan AI video generation model made by z. Where is sora Do you think that it will do much better https://t.co/Tw2EUW3iMk SUPER impressive FILM made by @TXhunyuan AI video generation model made by z. Where is sora Do you think that it will do much better https://t.co/Tw2EUW3iMk"  
[X Link](https://x.com/Xianbao_QIAN/status/1863856035850752404)  2024-12-03T08:01Z [----] followers, 25.4K engagements


"An open-source model is quickly catching up with @elevenlabsio Who is this 'Anonymous Sparkle' currently ranked second in the TTS rankings Its score is getting very close from ElevenLabs with so few votes"  
[X Link](https://x.com/Xianbao_QIAN/status/1864546908431372327)  2024-12-05T05:46Z [----] followers, 15.8K engagements


"The capabilities of vision-language models have soared this year with several now reaching GPT-4-level performance according to MMMU. And now theres a new addition to the family: Introducing InternVL [---] - Available in sizes ranging from 1B up to 78B parameters - Built upon the Intern-ViT encoder - Enhanced with chain-of-thought (CoT) reasoning for vision tasks - Code models and technical reports available"  
[X Link](https://x.com/anyuser/status/1865022396396376505)  2024-12-06T13:16Z [----] followers, [----] engagements


"CosyVoice [--] released by FunAudioLLM Team @TONGYI_SpeechAI - based on Qwen2.5-0.5B - added streaming support - support various Chinese dialects and role plays - allow language mixing Technical reports available. Models and demos will be available on @huggingface soon"  
[X Link](https://x.com/Xianbao_QIAN/status/1868508414487491041)  2024-12-16T04:08Z [----] followers, 13K engagements


"Infinigence open-sourced a multimodal model that handles image video and text all [--] modalities in one model. - 3B ideal for on-device deployment - SigLip-400M for vision understanding - Qwen2-Audio/whisper-large-v3 for audio understanding supporting both Chinese and English"  
[X Link](https://x.com/anyuser/status/1868563713407803863)  2024-12-16T07:48Z [----] followers, 10.6K engagements


"DeepSeek v3 appears to be performing deep thinking naturally and theres no longer a daily limit on deep thinking. Thats exciting However you can still only choose between online searching or deep thinkingjust one at a time. Probably due to context window limitation"  
[X Link](https://x.com/Xianbao_QIAN/status/1872137384223269064)  2024-12-26T04:28Z [----] followers, [----] engagements


"I've always felt that being bilingual isn't just about speaking two languages--it's about THINKING and muttering in whichever language feels more natural depending on the topic and context. For example I prefer doing math in Chinese because each digit is just one syllable which makes calculations crisp and efficient. But when it comes to topics like unconscious bias I automatically switch to English mainly because that's where I first learned and absorbed those ideas. It reminds me of programming languages: you can write almost any program in most languages but you still pick Bash for a quick"  
[X Link](https://x.com/Xianbao_QIAN/status/1878623350953857166)  2025-01-13T02:01Z [----] followers, 23.3K engagements


"Linear attention v.s. Transformer arch We have a new player in the OS LLM world. MiniMax a.k.a the company that released @Hailuo_AI Their first release is VERY impressive: - Linear attention with MoE - 456B total params with 45.9B activated params - 4M context length for inference - Both LLM VL and the base are available This would be an ideal model for agent use or long context understanding with great cost-efficiency and performance Demo & model links below"  
[X Link](https://x.com/Xianbao_QIAN/status/1879329052852736093)  2025-01-15T00:45Z [----] followers, [----] engagements


"InternLM v3 is live now - SoTA performance surpass models like Llama3.1-8B and Qwen2.5-7B - Capable of deep reasoning with system prompts (details in their model card) - Trained only on 4T high quality tokens. Energy saving"  
[X Link](https://x.com/anyuser/status/1879425413317001397)  2025-01-15T07:08Z [----] followers, 19.9K engagements


"DeepSeek r1 series is available on @huggingface now Surprises: - It's not just a single model it's a model family - Both big / small reasoning models available giving you more choices - r1 will be in MIT license allowing for model distillation - Released full training technics. Link below. Exciting"  
[X Link](https://x.com/anyuser/status/1881316371252015280)  2025-01-20T12:22Z [----] followers, 10.7K engagements


"Lets be clear: DeepSeek r1 isnt about who races fasterits about the inherent flaw in closed models. The future of AI isnt owned by those who hide code or stockpile chips. Its built on trust and trust requires transparency. When models are black boxes you surrender control over data privacy culturally aligned ethics and post-training customization for real world scenarios. Thats not leadershipits liability. Chinas pretraining consolidation proves a simple truth: AI is becoming a commodity. The real value lies not in the model but in what you do with it. Why waste billions reinventing"  
[X Link](https://x.com/Xianbao_QIAN/status/1884212238594236552)  2025-01-28T12:09Z [----] followers, 210.3K engagements


"The full demo is now available at @huggingface Check it out. Link below: Whoa Lumina-Image [---] is here 🔥 chinese community at it again 🏋 2B parameter model ⛓💥 Apache [---] license 🖼 Competitive to FLUXdev 8B (and it supports system prompts 🤯 ) https://t.co/Yi1B0eUoOb Whoa Lumina-Image [---] is here 🔥 chinese community at it again 🏋 2B parameter model ⛓💥 Apache [---] license 🖼 Competitive to FLUXdev 8B (and it supports system prompts 🤯 ) https://t.co/Yi1B0eUoOb"  
[X Link](https://x.com/anyuser/status/1888924354299502891)  2025-02-10T12:13Z [----] followers, 13.1K engagements


"People nailing Aha moment from @deepseek_ai R1 paper by running Open R1 on a surprisingly small 0.5B model. And it only need 4x4090 to work with proving you dont always need a massive compute setup to spark innovation. Pretty wild Open source verifiable rules could be big We just published the second OpenR1 update with OpenR1-220k-Math our new large-scale dataset for mathematical reasoning generated by DeepSeek R1. We generated 800k+ reasoning traces on [---] H100s in [--] days 🚀 https://t.co/odESKED0ms We just published the second OpenR1 update with OpenR1-220k-Math our new large-scale dataset"  
[X Link](https://x.com/Xianbao_QIAN/status/1889196451416412472)  2025-02-11T06:15Z [----] followers, 15.2K engagements


"@Alibaba_Qwen is on fire Just look at the results from @huggingface's Open LLM Leaderboard all top [--] models are Qwen derivatives That's amazing. Great work and HUGE thanks for delivering such great base model families that empower the community to innovate and build upon"  
[X Link](https://x.com/Xianbao_QIAN/status/1889551665235165540)  2025-02-12T05:46Z [----] followers, 57.7K engagements


"OpenVideo dataset is now available on @huggingface Literally a year ago shortly after Sora's release a group of passionate individuals set out to build a large-scale high-quality labeled dataset for video generation. I witnessed their journey from zeroovercoming countless challengesto successfully delivering [---] hours of labeled videos at 720p totaling 106K clips together with the tooling. Their dedication over an entire year was incredible. It was an honor to speak with them and see their commitment to an open-source approach and knowledge sharing. Link to their repo below:"  
[X Link](https://x.com/Xianbao_QIAN/status/1889643154112164202)  2025-02-12T11:50Z [----] followers, 14.2K engagements


"Who are the best GenAI companies in China Do you remember the Six Tigers (6 top unicorns) that were widely discussed in the mediaZhipu MiniMax Baichuan Kimi StepFun and [--] (note that this list excludes big players like Alibaba's Qwen) Two years later I believe that we now have the Six Open Source Tigers. Who are the major open source GenAI companies in China (again excluding the big ones like Alibaba and Tencent) Here is the list on top of my mind with one of their most impactful and representative models DeepSeek (@deepseek_ai) for r1 Zhipu (@ChatGLM) for CogVideoX OpenBMB (@OpenBMB) for"  
[X Link](https://x.com/anyuser/status/1891463272844644721)  2025-02-17T12:22Z [----] followers, 13.9K engagements


"The next DeepSeek is @StepFun_ai They just dropped two HUGE SoTA and commercial-grade models on @huggingface - T2V: 30B text to video model - Audio chat 130B audio understanding/generation model - mit/Apache [--] license Try them out before the servers get too crowded. Links:"  
[X Link](https://x.com/Xianbao_QIAN/status/1891525282948853950)  2025-02-17T16:29Z [----] followers, 37.5K engagements


"Moonshot the fated rival of DeepSeek finally joined the battle field of open source models. What a world"  
[X Link](https://x.com/anyuser/status/1893702111872844232)  2025-02-23T16:39Z [----] followers, 12.2K engagements


"Looking forward to Blibili's new TTS model"  
[X Link](https://x.com/Xianbao_QIAN/status/1894717575029985712)  2025-02-26T11:54Z [----] followers, [----] engagements


"A new virtual try-on app on @huggingface looks pretty cool"  
[X Link](https://x.com/anyuser/status/1900753054233989151)  2025-03-15T03:36Z [----] followers, 134.5K engagements


"Want to give more control on the 3D generation model New 3D generation model from @TXhunyuan is now available on @huggingface that supports multi-views Check it out: Hunyuan 3D MV Link below"  
[X Link](https://x.com/Xianbao_QIAN/status/1901834289966412177)  2025-03-18T03:13Z [----] followers, [----] engagements


"StepFun Ti2V just got released on @huggingface - 30B - up to [---] frames [--] s 540P video - Both text and image control - Precise motion control Link below"  
[X Link](https://x.com/Xianbao_QIAN/status/1902380130825310459)  2025-03-19T15:22Z [----] followers, 26.1K engagements


"Just realized that @Alibaba_Qwen is now an enterprise org on @huggingface :D Thanks for the support @JustinLin610 @cherry_cc12"  
[X Link](https://x.com/anyuser/status/1907561748699427298)  2025-04-02T22:32Z [----] followers, [----] engagements


"InternVL v3 is available now"  
[X Link](https://x.com/Xianbao_QIAN/status/1910659541836173599)  2025-04-11T11:41Z [----] followers, 12.2K engagements


"I also want to keep an eye on. Which one will come first 👀👀👀 https://t.co/mekr0Drodq https://t.co/7MQa2ak7oV 👀👀👀 https://t.co/mekr0Drodq https://t.co/7MQa2ak7oV"  
[X Link](https://x.com/anyuser/status/1916767538559795370)  2025-04-28T08:12Z [----] followers, 23.6K engagements


"A new reasoning model from Xiaomi Exciting"  
[X Link](https://x.com/anyuser/status/1917402568961241300)  2025-04-30T02:16Z [----] followers, 10.1K engagements


"Welcome the newest addition to the @Xiaomi MIMO family Remarkable multimodal reasoning power packed into a 7B model My take: The 7B scale is strategically important. Expect native execution on smartphones soon shifting your device's role from information repository to proactive loyal intelligent agent PS: the MiMo-VL-7B series maintains full compatibility with the Qwen2_5_VLForConditionalGeneration architecture for seamless deployment and inference. Best part You can deploy it right now Who will be the first @huggingface Inference Endpoint provider that supports this model"  
[X Link](https://x.com/anyuser/status/1928279488401658080)  2025-05-30T02:37Z [----] followers, 18.3K engagements


"RL training is too slow AReaL by @AntResearch_ introduces an asynchronous approach that decouples generation from training to eliminate blocking. Combined with system-level optimizations this method achieves up to [----] speedup. Code open-sourced @jxwuyi"  
[X Link](https://x.com/Xianbao_QIAN/status/1930083167756792189)  2025-06-04T02:04Z [----] followers, 30.4K engagements


"A new large audio language model from @StepFun_ai is now open sourced on @huggingface - Based on LLM 130B Step-Omni - Multi-stage training: pretrain SFT DPO model merging - Multilingual: Chinese (including Cantonese and Sichuanese) English Japanese etc"  
[X Link](https://x.com/Xianbao_QIAN/status/1933156502078112016)  2025-06-12T13:36Z [----] followers, 16.3K engagements


"2025.6.27 new LLM open sourced by @TencentHunyuan"  
[X Link](https://x.com/anyuser/status/1938520097016168775)  2025-06-27T08:49Z [----] followers, 34.1K engagements


"The first look on @TencentHunyuan's new LLM looks quite interesting: - Great performance for a 13B activation model (total 80B) - Both pretrain & instruct model released - Native quantization support for Int4 & FP8 via AngleSlim - 256K long context window with possibilities to future extend - Unified fast & slow (think-step) inference - Long context strong reasoning ready to be used by agents 2025.6.27 new LLM open sourced by @TencentHunyuan https://t.co/l34uaL09BV 2025.6.27 new LLM open sourced by @TencentHunyuan https://t.co/l34uaL09BV"  
[X Link](https://x.com/Xianbao_QIAN/status/1938525945352302734)  2025-06-27T09:12Z [----] followers, 33.6K engagements


"A company who had publicly denounced open source models being "stupid tax" has just re-embraced open source by releasing a series of models from 424B to 0.3B featuring both Paddle and Transformers support. Where are your models on @huggingface cc @OpenAI @sama @grok @elonmusk"  
[X Link](https://x.com/Xianbao_QIAN/status/1939503320286273954)  2025-06-30T01:56Z [----] followers, 34.3K engagements


"I'll never forget this model as well as the relationship between pretrain SFT and RLHF. Baidu just released [--] models at the same time on @huggingface - from 0.3B to 424B parameters. Lets go https://t.co/4cKj9VoWsZ Baidu just released [--] models at the same time on @huggingface - from 0.3B to 424B parameters. Lets go https://t.co/4cKj9VoWsZ"  
[X Link](https://x.com/Xianbao_QIAN/status/1939535258048929960)  2025-06-30T04:03Z [----] followers, 71.2K engagements


"How fast is LLM evolving GPT-4o last year is caught up by a 9B visual thinking model now. The new model from @thukeg seems to be performing extremely well on GUI agents. 9B fits well on laptops. I'm waiting for someone to develop a privacy first local browser agent on top of that. Link below:"  
[X Link](https://x.com/anyuser/status/1940244830191591614)  2025-07-02T03:03Z [----] followers, 35.1K engagements


"Oh. Is Kimi doc leaking their next generation model Probably the largest MoE that I've seen Kimi K2 1T total params 32B active params Super strong on coding & agent 128k context length no VL support ToolCalls JSON Mode search https://platform.moonshot.cn/docs/pricing/chat#%E7%94%9F%E6%88%90%E6%A8%A1%E5%9E%8B-kimi-k2 https://platform.moonshot.cn/docs/pricing/chat#%E7%94%9F%E6%88%90%E6%A8%A1%E5%9E%8B-kimi-k2"  
[X Link](https://x.com/Xianbao_QIAN/status/1943621126652821617)  2025-07-11T10:39Z [----] followers, 67.3K engagements


"Kimi K2 is open sourced on @huggingface - 1T MoE 32B active params - Excellent coding & Tool use & Math - Not a thinking model - Both BASE and Instruct is released friendly for fine-tunes https://huggingface.co/moonshotai/Kimi-K2-Base Oh. Is Kimi doc leaking their next generation model Probably the largest MoE that I've seen Kimi K2 1T total params 32B active params Super strong on coding & agent 128k context length no VL support ToolCalls JSON Mode search https://t.co/OxvyodOQbI https://t.co/2cJrILXXln https://huggingface.co/moonshotai/Kimi-K2-Base Oh. Is Kimi doc leaking their next"  
[X Link](https://x.com/Xianbao_QIAN/status/1943685911851479388)  2025-07-11T14:56Z [----] followers, 29.3K engagements


"The value of R2: Open source model is catching up with frontier model on agentic tool use. They're serious. Why Here are the details: I used to believe that we at least need a thinking model for agentic tool use case. which is a bad news to be honest because it will waste a ton of tokens and make the agent super slow. Kimi K2 told me that I'm totally wrong. A non-thinking model can do it very well. K2 is probably the best open source agentic tool use model on par with top close source models. I connected with Claude code and ask it to write projects for me involving complicated planning and"  
[X Link](https://x.com/anyuser/status/1943696007877009451)  2025-07-11T15:36Z [----] followers, 14.3K engagements


"Sry for spamming the timeline. I was just too excited with K2 release. It reminds me of R1 launch all over again the first time open-source models caught up to the closed-source frontier After R1 release Gemini [---] Pro ChatGPT-o3 and Claude [--] surged ahead and R1 can't keep pace with them anymore after [--] months putting open-source performance right back on the frontiers heels and how K2 arrives we have an open source models with performance on par with frontier models again what will happen I don't know. But this is big and beautiful no"  
[X Link](https://x.com/Xianbao_QIAN/status/1943708350656684286)  2025-07-11T16:25Z [----] followers, 13.4K engagements


"Another reason on why this is big. from my test K2 performs super well on coding and it works seamlessly with Claude code and other agentic coding apps Both Claude and ChatGPT has very tight geolocation policy meaning that a lot of Chinese developers are unable to use them even if they are on VPNs And now they can just flew to Kimi meaning that the adoption rate of K2 could be significant (if everything works well) (i.e. they had enough compute great SLA) (and K2 API offers a reasonable price) The value of R2: Open source model is catching up with frontier model on agentic tool use. They're"  
[X Link](https://x.com/anyuser/status/1943729056018837880)  2025-07-11T17:48Z [----] followers, [----] engagements


"Why did @Kimi_Moonshot switched from closed source to open-source and released K2 - Reputation. If K2 is API only it might have ended up like @grok-4 - clearly well-built yet still taking a lot of flak. - Work with the whole ecosystem. Less than [--] hours after release the community had already shipped an MLX port 4-bit quantization and more - things that the small Kimi team doesn't have the manpower to do. - But the most important one is to a higher technical bar and forces the researchers to build a better model - You can't just claim to have a great model you need to prove it by allowing"  
[X Link](https://x.com/anyuser/status/1944442400619089983)  2025-07-13T17:02Z [----] followers, 21.5K engagements


"Seed Prover solved [--] out of [--] IMO questions in [--] days and got Silver. Proof: Big congratulations to @huajian_xin Now you know what I'm going to kindly ask: Would you consider open sourcing it :D https://github.com/ByteDance-Seed/Seed-Prover/tree/main/SeedProver/imo2025 https://github.com/ByteDance-Seed/Seed-Prover/tree/main/SeedProver/imo2025"  
[X Link](https://x.com/Xianbao_QIAN/status/1947895409600565397)  2025-07-23T05:43Z [----] followers, 21.9K engagements


"Intern-S1 a new multimodal model from @intern_lm - 235B MoE + 6B vision encoder - 5T multimodal tokens & 2.5T scientific-domain tokens - great model for AI4S research - support tool calling capabilities Model on @huggingface: https://huggingface.co/internlm/Intern-S1 https://huggingface.co/internlm/Intern-S1"  
[X Link](https://x.com/anyuser/status/1949123899645714657)  2025-07-26T15:05Z [----] followers, 24.1K engagements


"18+ researchers are waiting. GLM [---] was a big surprised. What can we get from GLM-4.5 🤩"  
[X Link](https://x.com/anyuser/status/1949726918867103887)  2025-07-28T07:01Z [----] followers, [----] engagements


"Step [--] has just been released. It proposed a new infra level optimization of Attention FFN disaggregation. Model & Infra co-design is the way forward Model: Technical paper: https://arxiv.org/abs/2507.19427 https://huggingface.co/stepfun-ai/step3 https://arxiv.org/abs/2507.19427 https://huggingface.co/stepfun-ai/step3"  
[X Link](https://x.com/Xianbao_QIAN/status/1950917239915397334)  2025-07-31T13:51Z [----] followers, 19.1K engagements


"The new talking head model EchoMimicV3 from Ant Group seems to be pretty cool. Based on Wan [---] 1.3B"  
[X Link](https://x.com/anyuser/status/1954955197807313192)  2025-08-11T17:16Z [----] followers, 26K engagements


"Very impressive multimodal understanding model from @Zai_org - 106B A12B model - MIT license model weights - Supports grounding - Able to handle GUI tasks - Image/video understanding & long doc parsing"  
[X Link](https://x.com/anyuser/status/1955252600066212185)  2025-08-12T12:58Z [----] followers, 14.3K engagements


"A very interesting food dish dataset if you're building a health app/model: [---] k carefully curated food samples spanning home-cooked meals restaurant dishes raw ingredients and packaged products. How it was built is just as valuable [--] k real users on Binance captured their own plates and pre-annotated by professional human annotators. Machine-generated labels were then spot-checked and refined by Biance users to guarantee quality. A slice of the dataset made available on Hugging Face under an OpenRail license. Sounds like a new approach for crowdsourcing data collection. Link below:"  
[X Link](https://x.com/Xianbao_QIAN/status/1955935971385729431)  2025-08-14T10:14Z [----] followers, 91.4K engagements


"A fully open-sourced top tier Deep Research framework. Guess which one it is"  
[X Link](https://x.com/anyuser/status/1955973368039268852)  2025-08-14T12:42Z [----] followers, 38.9K engagements


"ToonComposer: You can now efficiently make cartoons on @huggingface for free - Input: sketch based key frames + color reference frame - This @Alibaba_Wan based model will combine in-betweening & colorization - Model can also imagine areas left blank with a prompt - Result: save up to 70% of manual work. Huge thanks to B&T Studio and Gudong Animation Studio for their permission to use their animation content (Big Fish & Begonia and Mr. Miao) for academic illustration"  
[X Link](https://x.com/Xianbao_QIAN/status/1957325826778914828)  2025-08-18T06:16Z [----] followers, 43.4K engagements


"nano-banana qwen-image-edit what else Try @StepFun_ai NextStep-1-Large-Edit - 14B AR model - Apache [--] license - Demo available on @huggingface - Pretrain model also made available Link below"  
[X Link](https://x.com/Xianbao_QIAN/status/1957749693485838448)  2025-08-19T10:21Z [----] followers, 51.9K engagements


"I saw some screenshot on wechat that the new upcoming AutoGLM from @Zai_org is mind blowing that you can even play video games inside. Just imagine one day all our computer usage is via virtual machine like @e2b. Computers can be made for free as long as you agree with donating all your computer usage data for training"  
[X Link](https://x.com/Xianbao_QIAN/status/1958029177564934349)  2025-08-20T04:51Z [----] followers, 18.9K engagements


"ByteDance Seed OSS is now available on @huggingface - 36B Apache2 - Great for ablation research including Base base without synthetic instruction data and instruction variants - Great performance on reasoning & Agents - Native 512k long context - Flexible thinking budget @Xianbao_QIAN model release now github: https://t.co/SLUuitdGVj huggingface model:https://t.co/8Lu1oJK6C7 @Xianbao_QIAN model release now github: https://t.co/SLUuitdGVj huggingface model:https://t.co/8Lu1oJK6C7"  
[X Link](https://x.com/anyuser/status/1958344544531235294)  2025-08-21T01:44Z [----] followers, 32K engagements


"Meituan just open sourced their new MoE LLM LongCat on @huggingface It's exciting to see new players The model looks very interesting too with technical report. https://huggingface.co/meituan-longcat/LongCat-Flash-Chat https://huggingface.co/meituan-longcat/LongCat-Flash-Chat"  
[X Link](https://x.com/Xianbao_QIAN/status/1961917378968899986)  2025-08-30T22:22Z [----] followers, 106.4K engagements


"I'm very glad to see that the new translation model from @TencentHunyuan is now ranking the 3rd. It's a reminder that small domain tuned models are more valuable than they appears. Agentic stack needs both large and small models. Large models can handle planning and leverage sub-agents based on lean models to perform a particular task. Small models are cheap fast and fine-tunable. They're not the opposite of large models but the complement to it"  
[X Link](https://x.com/anyuser/status/1963158909407109623)  2025-09-03T08:35Z [----] followers, 20.4K engagements


"Actually I underestimate and now super bullish on @Baidu_Inc Did you quietly train a VL model using your own Kunlun P800 chips instead of Nvidia GPUs and not telling the world about it That's BIG Looks like the money cloud companies sent on chips is starting to pay off @baidu is back Two models and a new global web portal has been released today. Finally the phone verification is no longer needed They do listen to the voice of developers. This model release includes both enhanced lightweighted thinking model ERNIE-4.5-21B-A3B-Thinking https://t.co/siU2s4rtQ3 @baidu is back Two models and a"  
[X Link](https://x.com/anyuser/status/1966443608858402869)  2025-09-12T10:07Z [----] followers, 14.9K engagements


"WAN [---] animate model & demo is now officially released on @huggingface"  
[X Link](https://x.com/Xianbao_QIAN/status/1968882983672365533)  2025-09-19T03:40Z [----] followers, 51.7K engagements


"New @deepseek_ai models coming"  
[X Link](https://x.com/anyuser/status/1970095834978242983)  2025-09-22T12:00Z [----] followers, 36.9K engagements


"I came across AIWood's review on @TencentHunyuan Image [---]. It's an amazing model with 80B weights. Big = Better quality. No wonder Nano banana is just Gemini [---] Flash Image"  
[X Link](https://x.com/anyuser/status/1971577053872099791)  2025-09-26T14:06Z [----] followers, 47.3K engagements


"The largest 80B open source image generation model has been dropped on @huggingface Review video by AIWood below. I came across AIWood's review on @TencentHunyuan Image [---]. It's an amazing model with 80B weights. Big = Better quality. No wonder Nano banana is just Gemini [---] Flash Image. https://t.co/DBU851kW3m I came across AIWood's review on @TencentHunyuan Image [---]. It's an amazing model with 80B weights. Big = Better quality. No wonder Nano banana is just Gemini [---] Flash Image. https://t.co/DBU851kW3m"  
[X Link](https://x.com/anyuser/status/1972221107181883587)  2025-09-28T08:45Z [----] followers, 37.2K engagements


"MinerU [---] has arrived with demo on @huggingface"  
[X Link](https://x.com/anyuser/status/1972551152983228582)  2025-09-29T06:36Z [----] followers, 162K engagements


"PaddleOCR-VL-0.9B is mind blowing and it supports [---] languages Check it out on HF demo:"  
[X Link](https://x.com/Xianbao_QIAN/status/1978801182400475402)  2025-10-16T12:32Z [----] followers, 29.7K engagements


"China's open source is just on fire. Soul China's Tinder () has just open sourced their podcast model on @huggingface https://huggingface.co/collections/Soul-AILab/soulx-podcast https://huggingface.co/collections/Soul-AILab/soulx-podcast"  
[X Link](https://x.com/Xianbao_QIAN/status/1983429688606540141)  2025-10-29T07:04Z [----] followers, 279.7K engagements


"This work from @BytedanceTalk seed team will be transformative and open the era of iterative latent reasoning: why do model have to only think in human languages At least I don't. The result is also significant: - 2.6B R4 (4 steps) model achieved comparable performance with qwen3 8B and Gemma [--] 12B The project was supervised by @Yoshua_Bengio @jasoneshraghian How do you feel about this work"  
[X Link](https://x.com/Xianbao_QIAN/status/1985608034778030106)  2025-11-04T07:20Z [----] followers, 22.3K engagements


"The same day U.S. humanoid startup KScale shuts down @XPengMotors a major EV maker released their new generation humanoid robot IRON It's mind-blowing: - Targeted mass production: [----] - 3x XPeng home made Turning AI chip with [----] TOPS - Human-like spine with full cover soft skin - Powered by Physical World Large Model - Early user pilot in industrial areas Is XPeng going to be the next TSLA + NVDA Potentially time to long NYSE: XPEV. K-scale cancels orders and refunds deposits for kbot. I thought all the VCs were excited about US-based robotics what happen https://t.co/eVKhoypkHq K-scale"  
[X Link](https://x.com/Xianbao_QIAN/status/1986051187519475779)  2025-11-05T12:41Z [----] followers, 174.6K engagements


"Still skeptical about @XPengMotors latest robot "Theres definitely a human hiding in there" Before jumping to conclusions lets take a rare look inside. XPengs founder Xiaopeng just pulled back the curtain by revealing what actually powers this machine. This footage might surprise you @teortaxesTex Show me any footage where you can see the robot move like this outside of this live demo. Any footage where you can see the internals. Demos where it moves with this fluidity. There isn't any. Because it's a human in a robot suit. @teortaxesTex Show me any footage where you can see the robot move"  
[X Link](https://x.com/Xianbao_QIAN/status/1986336248651477461)  2025-11-06T07:34Z [----] followers, 10.8K engagements


"Kimi-k2-thinking This is real"  
[X Link](https://x.com/Xianbao_QIAN/status/1986368577587618091)  2025-11-06T09:42Z [----] followers, 16.3K engagements


"Love this @Zai_org posted "Awesome" under @Kimi_Moonshot's recent release post. In open source we join hands and collaborate"  
[X Link](https://x.com/Xianbao_QIAN/status/1986629297562648636)  2025-11-07T02:58Z [----] followers, [----] engagements


"The new Kimi model was trained with only $4.6M Like 1/100 of their US counterparts Can that be true Remember: In [----] NASAs Apollo mission landed people on the moon with a computer that had just 4KB of RAM. And as @crystalsssup said "Creativity loves constraints" If you ever wonder how Chinese frontier models like Kimi DeepSeek and Qwen are trained on far fewer (and nerfed) Nvidia GPUs than US models. In [----] NASAs Apollo mission landed people on the moon with a computer that had just 4KB of RAM. Creativity loves constraints. If you ever wonder how Chinese frontier models like Kimi DeepSeek"  
[X Link](https://x.com/Xianbao_QIAN/status/1986744488623317143)  2025-11-07T10:36Z [----] followers, 338.4K engagements


"wait.what I knew it was good but I didn't know that it was this good 😍 MoonshotAI has released Kimi K2 Thinking a new reasoning variant of Kimi K2 that achieves #1 in the Tau2 Bench Telecom agentic benchmark and is potentially the new leading open weights model Kimi K2 Thinking is one of the largest open weights models ever at 1T total parameters https://t.co/uMy32XhdiG MoonshotAI has released Kimi K2 Thinking a new reasoning variant of Kimi K2 that achieves #1 in the Tau2 Bench Telecom agentic benchmark and is potentially the new leading open weights model Kimi K2 Thinking is one of the"  
[X Link](https://x.com/anyuser/status/1986948577567100993)  2025-11-08T00:07Z [----] followers, 12.5K engagements


"Welcome Weibo to the open source AI community They just released a 1.5B reasoning model that surpasses DeepSeek R1-0120 on math while being 100x smaller. Model: https://huggingface.co/WeiboAI/VibeThinker-1.5B VibeThinker-1.5B SOTA reasoning in a tiny model. 🚀 Performance: Highly competitive on AIME24/25 & HMMT25 surpasses DeepSeek R1-0120 on math and outperforms same-size models in competitive coding. ⚡ Efficiency: Only 1.5B params 100-600 smaller than giants like Kimi K2 https://t.co/cL2kaZ7uIs https://huggingface.co/WeiboAI/VibeThinker-1.5B VibeThinker-1.5B SOTA reasoning in a tiny model."  
[X Link](https://x.com/Xianbao_QIAN/status/1988456123469635628)  2025-11-12T03:57Z [----] followers, 49.1K engagements


"@miromind_ai released their open source latest model on research agent. - interleaved thinking with multi-step analysis powered by RL - 256k context window - handles up to [---] tool calls per task - comes in 8B 30B and 72B for different compute budgets - MiT license Model available on @huggingface demo and technical report ready"  
[X Link](https://x.com/Xianbao_QIAN/status/1988929067203326322)  2025-11-13T11:16Z [----] followers, 59.9K engagements

Limited data mode. Full metrics available with subscription: lunarcrush.com/pricing