Pareto Matches Top Models on 105-Bug Benchmark for $4.81

A new Bug Hunt Bench evaluation planted 105 real bugs across two code repositories, finding that Unbiased's Pareto composite model nearly matches GPT-6 Astra-class models at just $4.81 in cost.

2026-09-18 ~ 2026-09-18 · 2 related posts

• 105 planted bugs benchmark: Unbiased's Pareto scores 30.7 for just $4.81 — PawelHuryn · 2026-09-18
• Bug Hunt Bench: 105 real planted bugs, GPT-6 Astra leads at 45, Pareto hits 30.7 for just $4.81 — PawelHuryn · 2026-09-18