banteg's Sokoban Benchmark: GPT-6.2 Models Clear All Levels
Developer banteg launched a Sokoban reasoning benchmark of 66 puzzle levels for LLMs; both gpt-6.2-sol and gpt-6.2-astra cleared all levels, finishing just an hour apart, with gpt-6.2-sol completing the run at about 1/6.5 the cost of Fable 5.2.
2026-11-18 ~ 2026-11-18 · 3 related posts
- GPT-6.2 Races Fable 5.2 in Live Sokoban Puzzle Eval, Trails by Just One Hour — banteg · 2026-11-18
- Sokoban eval for LLMs: gpt-6.2 sol lands #2, astra #3; humans invited to beat model routes — banteg · 2026-11-18
- GPT-6.2 Sol clears Sokoban eval at 1/6.5th the cost of Claude Fable — banteg · 2026-11-18