Bohan Zhai

🎉 CaptionQA is accepted to CVPR 2026! If you care about captioning in real systems, we built CaptionQA to be simple and practical. 📄 Paper: arxiv.org/pdf/2511.21025… 🙌 Welcome to try CaptionQA on your models and share results! #CVPR2026 #MLLM #Benchmark #Captioning

Let’s chat @ COLM!

I thrive on deep research, uncovering the insights that lie behind impressive benchmark scores. As I often discussed with my friends and collaborators, while achieving high scores can attract immediate attention, it's a short-term win. True progress requires understanding the science behind what works. Without this, we eventually hit limitations. This is why lots of companies, despite significant investment, still find themselves lagging behind leaders like OpenAI and Claude.

🚀 Introducing our new paper: "The Law of Vision Representation in MLLMs"! ArXiv: arxiv.org/abs/2408.16357 GitHub: github.com/bronyayang/Law… Big thanks to our amazing team: @Chenfeng_X, @BohanZhai, and other coauthors. (1/4)