GPT-6.2 Sol hits 54.7 on Bug Hunt for $7, matching Fable 5.2 at 6% of the cost
PawelHuryn · x · 2026-11-18
Pawel Huryn posted GPT-6.2 Sol results on Bug Hunt Bench (planted bugs only, n=3 per model):
- max: 54.7/105, 68 min, $7.00
- xhigh: 49/105, 41 min, only $3.89
- vs GPT-6.1 Sol: +10.4 points at max, +6.3 at xhigh
- Sol (xhigh) ties Fable 5.2 (xhigh) at 49 for 6% of the cost
- Leaderboard top: GPT-6.2 Astra Max at 62.7; Sonnet 5.5 Max scores 51.3 but takes 287 min and $154
- DeepSeek V4.1 Flash and Kimi K3.1 Pro are the only real-billing models listed
More effort levels for GPT-6.2 Astra and Sol coming; he also shares free agent-teams skills for running Astra as a review subagent from Opus 5.5.
More from coding & agent
- Record your judgment process 3 times, let AI build business evals from it — alliekmiller · 2026-11-18
- Noodle: Free open-source Mac app runs collaborative personal AI agents locally — Dave_Maynor · 2026-11-18
- Doing EQ/compression on voiceovers with FFmpeg inside Antigravity — alexanderchen · 2026-11-18
- A 10,000-agent swarm solved NS — a single Codex agent is not the same thing — daniel_mac8 · 2026-11-18
- VulKill launches hosted MCP server for code review inside your coding client — VulKill · 2026-11-18
- Screenshot shows Amazon blocking AI agents with 'Access restricted' warning — Dave_Maynor · 2026-11-18