$500 RL fine-tuned 9B model outperforms Frontier

A hobbyist/independent experiment shows a $500 reinforcement-learning fine-tune on a 9B-parameter open model beat Frontier's models on a catalog-review task, signaling that smaller, cheaper open models plus modest RL training can rival larger proprietary offerings. This suggests potential pressure on pricing and differentiation for firms selling larger commercial models and may accelerate adoption of lightweight open-model workflows in some enterprise tasks.