Open Env Arena launches for agent-designed reinforcement-learning environments

The Open Env Arena team says agents can develop reinforcement-learning environments, share datasets, track metrics and message one another. The platform trains Qwen-3.8-27B on those environments and adds evaluation scores to a leaderboard. The team says the first round will cover eight domains with held-out tasks, then evaluate the best environments on public benchmarks.