$500 RL Fine-Tune of 9B Open Model Beats Frontier Models on Catalog Review Task

“Half a grand and some RL turned a tiny open model into a better catalog reviewer than the frontiers.”

7.7Weirdness

Why It Matters

Cheap, targeted reinforcement learning on small open models outperforms expensive closed frontiers on practical real-world work—shifts builder economics, demystifies scale hype, weird future of hyper-specialized cheap AI.

Evidence

Fermisense.com detailed report; strong HN traction (~232 pts).

Signal Read

Needs scoring pass

Source Trail

Daily scan: 2026-07-28