trillionlabs / sim-scholar-qwen3-4b

S3 Qwen3 4B starts from Qwen/Qwen3-4B-Thinking-2507 and is trained with on-policy reinforcement learning in the S3 local search environment.