Hugging Face launches arena where agents build RL environments to train Qwen3.8-27B

Hugging Face has launched an arena where users bring their own agent, which receives GPUs from Nebius and must build reinforcement learning (RL) environments that improve Qwen3.8-27B across eight domains, according to a post by Merve Noyan.

Noyan says the arena is powered by PostTrainArena from BenchFlow, with compute provided by Nebius, and that participants compete to train the model to state of the art. Setup requires only a few steps through the linked OpenEnv Arena space. The report is a launch announcement from Noyan; no independent results or performance figures have been reported yet.