GPT-6.2 Astra Tops Bug-Hunt Coding Benchmark at 62.7/105, One-Fifth Sonnet's Cost
PawelHuryn · x · 2026-11-18
Pawel Huryn ran the new models through Bug Hunt Bench, a real-world test of finding and fixing 105 planted bugs across two repos. GPT-6.2 Astra takes the top spot:
- GPT-6.2 Astra (max): 62.7/105 at $28.35
- GPT-6.2 Astra (xhigh): 60/105 at $23.53
- Fable 5.2 (max): 58/105 at $139.65
- Gemini 4 Carbon (high): 57.3/105 at $31.32
- Sonnet 5.5 (max): 51.3/105 at $154.03
- Previous GPT-6 Astra (max): 45/105 at $33.03
That's a +17.7-point jump over GPT-6 Astra at max effort, at roughly one-fifth of Sonnet 5.5 max's cost. The full leaderboard also shows results for DeepSeek V4.1 Flash, Kimi K3.1 Pro and more. His verdict: whatever people say about other use cases, it's a code review beast.
More from coding & agent
- Dev wants a routing model to dispatch intents between Codex and Claude MCPs — TheZachMueller · 2026-11-18
- One dev ships full zombie survival game built by a team of Hermes AI agents — Teknium · 2026-11-18
- 10 ChatGPT + Codex prompts to diagnose and fix a slow laptop yourself — anthara_ai · 2026-11-18
- Agent auto-iterates Triton kernels in background, beats all OSS implementations on Strix Halo — qtnx_ · 2026-11-18
- SimplePost Gives ChatGPT and Claude a Publishing Button for 10 Social Networks — haltakov · 2026-11-18
- Obsidian Starter Kit gives Claude Code, Codex and Gemini the same persistent context — dSebastien · 2026-11-18