RL-Trained 35B Open Model Cuts Code Search Costs 100x

Applied Compute and turbopuffer showed that an RL post-trained 35B open model can beat frontier-model code search over prebuilt indexes of 9,000 GitHub repos at roughly 1/100 of the cost.

2026-09-05 ~ 2026-09-05 · 2 related posts

• Post-Training a 35B Open Model for Code Search: 100x Cheaper Than Frontier Models — ypatil125 · 2026-09-05
• 35B Open-Weight Model Tops Code Search at 100x Lower Cost Than Frontier Models — ypatil125 · 2026-09-05