RAX robot agent adds DeepSeek V4.1 Flash: +10% over π0.5 on DexJoCo at 1/20 Opus 5.5 cost
RAX adds DeepSeek Harness and self-hosted V4.1 Flash support; with just 3 demonstrations it beats π0.5 and Fast-WAM by over 10% on DexJoCo at roughly 1/20 of Opus 5.5 API cost.
🚀 RAX now supports DeepSeek Harness and Kimi Code as agent harnesses! We’ve also added direct support for self-hosted DeepSeek-V4.1-Flash, so you can run RAX with local inference. On the DexJoCo benchmark, RAX + DeepSeek-V4.1-Flash outperforms π0.5 and Fast-WAM by more than 10% with just 3 demonstrations. At API pricing, it costs roughly 1/20 as much as Claude Opus 5.5! Try it out! 🤖 🌐 Project + sessions: https://robot-agent-exploration.com 📄 Paper: https://arxiv.org/abs/2610.22496 💻 GitHub: https://github.com/robot-agent-exploration/rax 🧩 Skill library: https://github.com/robot-agent-exploration/skills-lib 🤗 Dataset with full robot logs and sessions: https://huggingface.co/datasets/robot-agent-exploration/robot-session-logs @deepseek_ai @Kimi_Moonshot #Robotics #EmbodiedAI #AIAgents #OpenSource #DeepSeek #RobotLearning #LLM
This post quotes another post: x.com/du_yilun/status/2111143189222404427