Fermisense

A $500 RL fine-tune of a 9B open model beat frontier models on catalog review

A blog post (surfaced via Hacker News with 182 points) claims that a reinforcement learning fine-tune of a 9B open-weights model costing approximately $500 outperformed frontier models on a catalog review task. The result, if reproducible, illustrates that task-specific RL fine-tuning can close or reverse the gap with much larger proprietary models at very low cost. The post attracted meaningful community engagement with 48 comments.