GLM 5.1 from @Zai_org ranks as the top open model on the newly released Monthly-SWEBench by @UniPat_AI—second only to Claude-Opus-4.6. Congrats to the team!
🚀Explore the benchmark: unipat.ai/benchmarks/Mon…
English
Liang Chen
276 posts
Liang Chen
@liangchen5518
Cofounder of @UniPat_AI. I worked at Moonshot AI, Alibaba Qwen and Microsoft Research Asia.
Beijing Katılım Şubat 2022
203 Takip Edilen3K Takipçiler
Liang Chen retweetledi
Great progress on BabyVision!
Monishwaran Maheswaran@sudomonish
Super excited to Introduce our latest work: Squeeze Evolve. We unify test-time scaling methods into one evolutionary framework — then orchestrate many models across it. 3x lower cost. 10x throughput. 97.5%(SoTA) on ARC-AGI-V2. No verifier required. Framework: squeeze-evolve.github.io
English
@gietema @JRobertsAI @anshsharma009 @SamuelAlbanie @UniPat_AI Really nice blog about the VLM evaluations!
English
@JRobertsAI @anshsharma009 @SamuelAlbanie BabyVision by @liangchen5518 and others from @UniPat_AI is a good dataset as well - especially given that they actually reported the performance of 3, 6, 10 and 12-year-olds on (a subset of) this dataset
English
New blog post: an overview of what I learnt by browsing 28 VLM eval datasets.
English
2026年了还在做这个呢 早说了 vision token 会传播到 text token 里被后续 token readout
-Zho-@ZHO_ZHO_ZHO
啊？ 多模态模型有时根本没看图，却答得像看了一样真？ 斯坦福最新研究指出：多模态模型存在 “mirage reasoning（海市蜃楼式推理）”，即模型根本没看/有图像，照样生成详细图像描述和推理，甚至在医学 benchmark 上拿高分 不少多模态其实并非真正视觉理解，论文因此提出了更好的评估方案：B-Clean
中文
In 2026, we’re seeing intelligence translate more directly into real economic impact. Echo is a distinctive example of this—creating value by predicting the future.
Lucky to be building with such a great team @UniPat_AI.
UniPat AI@UniPat_AI
Today we’re introducing Echo — our full-stack prediction intelligence system, which turns uncertainty🔮 into profit📈. We Make Prediction General, Evaluable, Trainable and Profitable. 🌐Website: echo.unipat.ai
English
The scope of CUA (Computer Use Agent) should be extended more than GUI agent. What do you think CUA is?
English
Liang Chen retweetledi
Also check our opensource code here! github.com/UniPat-AI/SWE-…
English
Introduce SWE-Vision: A Minimal Agent for Advancing Visual Intelligence
While LLMs' coding have surpassed human performance in many benchmarks, visual reasoning still lags behind. We apply a simple stateful env and reach the sota perf.
The blog in 👉 unipat.ai/blog/SWE-Vision
English
Liang Chen retweetledi
We have taken a huge step forward in computer use.
:cua-sleep-rave-repeat:
OpenAI@OpenAI
GPT-5.4 Thinking and GPT-5.4 Pro are rolling out now in ChatGPT. GPT-5.4 is also now available in the API and Codex. GPT-5.4 brings our advances in reasoning, coding, and agentic workflows into one frontier model.
English
Congrats Qwen👑! Already feeling the acceleration of visual language intelligence in 2026 — looking forward to the “adult vision” moment.
UniPat AI@UniPat_AI
🔥 BabyVision Leaderboard Update Qwen3.5-397B-A17B now ranks as the #1 open-source model on the BabyVision Benchmark — scoring 43.3 without tools. Huge congrats to the team @Alibaba_Qwen! Check the full leaderboard from unipat.ai/benchmarks/Bab…
English
Liang Chen retweetledi
🚀 Qwen3.5-397B-A17B is here: The first open-weight model in the Qwen3.5 series.
🖼️Native multimodal. Trained for real-world agents.
✨Powered by hybrid linear attention + sparse MoE and large-scale RL environment scaling.
⚡8.6x–19.0x decoding throughput vs Qwen3-Max
🌍201 languages & dialects
📜Apache2.0 licensed
🔗Dive in:
GitHub: github.com/QwenLM/Qwen3.5
Chat: chat.qwen.ai
API：modelstudio.console.alibabacloud.com/ap-southeast-1…
Qwen Code: github.com/QwenLM/qwen-co…
Hugging Face: huggingface.co/collections/Qw…
ModelScope: modelscope.cn/collections/Qw…
blog: qwen.ai/blog?id=qwen3.5
English
Liang Chen retweetledi
NEW SOTA 👑
Seed-2.0 Pro is now the #1 model (60.6) on the BabyVision Benchmark, overtaking Gemini-3 Pro (49.7).
Huge congrats to the #BytedanceSeed team!
English
Liang Chen retweetledi
BabyVision
Visual Reasoning Beyond Language
huggingface.co/papers/2601.06…
English
Excited to introduce BabyVision, a benchmark at the “beginning” of human visual intelligence yet challenging to frontier multimodal LLMs!
UniPat AI@UniPat_AI
Can frontier MLLMs see like a 3-year-old? We’re releasing BabyVision — a vision-centric benchmark that isolates pre-linguistic visual primitives kids solve effortlessly, but models still struggle with.👇
English
Despite Gemini 3's superhuman performance in various domains, accuracy in basic visual tasks like cube counting is still a work in progress. Advancement in VLM seems to be slower than that of LLM. New pattern and paradigm are needed. gemini.google.com/share/a18f65dc…
English