Kaituo Feng

main experiment results

🔥 Introducing OpenSearch-VL: a fully open recipe for training SOTA-level multimodal search agents! 📦 Releases data, code, models, and data pipeline. 🏆 Achieves about 15-point gains across 7 benchmarks, including BC-VL, FVQA, SimpleVQA. 📜 Paper: arxiv.org/pdf/2605.05185

data pipeline

solid and exciting work！

@purshow04 Thanks for your interest!

@KaituoFeng Impressive work! I am certain that using search agents to assist in generation will be a popular paradigm for aigc model products. Thank you for using our WISE!

🔥 Introducing Gen-Searcher: the first attempt to train a deep research agent for image generation! 🏆 Achieves 15+ point gains on KnowGen and WISE! 🚀 All SFT & RL data with code are released. 📜 Paper: arxiv.org/pdf/2603.28767 🌍 Page: gen-searcher.vercel.app

@ApryleWu lol

@KaituoFeng uzi 又夺冠了😭

@_akhaliq Our Project page: gen-searcher.vercel.app Code: github.com/tulerfeng/Gen-… Model and Data: huggingface.co/GenSearcher

Gen-Searcher Reinforcing Agentic Search for Image Generation paper: huggingface.co/papers/2603.28…

@NirDiamantAI @_akhaliq Thanks for sharing this insight! That's a good idea.

@_akhaliq btw the search space pruning happens at token level, so you can actually cache intermediate search states between generations

@Chain_AlphaX @_akhaliq Thanks!

@_akhaliq ooh, interesting stuff. gonna check that out later.

@Elon_Orbit @_akhaliq Thanks! Yes — our method uses multi-hop search and reasoning, trained through a dual-reward agentic RL framework.

@_akhaliq Finally tackling frozen knowledge in image gen. Multi-hop reasoning + dual reward RL? Clever architecture. Full open-source earns real respect.

@_akhaliq Huge thanks for sharing our work!

Visualization comparison

Performance on WISE

Performance on KnowGen

Our data pipeline

Our inference process demo

@_akhaliq Thank you for sharing our work! Our full project can be found here: github.com/tulerfeng/Vide…

discuss: huggingface.co/papers/2503.21…

Video-R1 Reinforcing Video Reasoning in MLLMs

@_akhaliq Thank you for sharing our work! Our full project can be found here: github.com/tulerfeng/Vide…