On: VibeThinker: 3B param model that beats Opus 4.5 on reasoning with novel SFT+GRPO

[nolist_policy]

Notable:

  VibeThinker-3B is developed through a staged post-training pipeline built upon Qwen2.5-Coder-3B base, a compact 3B foundation model.

Qwen2.5 is ancient by LLM standards.