GPT-6.2 Sol clears Sokoban eval at 1/6.5th the cost of Claude Fable
banteg · x · 2026-11-18
Developer banteg ran a Sokoban eval (66/66 rooms) across frontier models: GPT-6.2 Sol and GPT-6.2 Astra (medium effort) both cleared it, landing #2 and #3 behind Claude Fable 5.2, with Claude Opus 5.5 fourth. The cost gap is stark — Fable took 3:27:12 at $51.88 and Opus $53.63, while Sol finished slightly slower (3:39:46) at just $7.93 (6.5x cheaper than Fable) and Astra at $19.86. Replays available for step-by-step comparison.
Related event: banteg's Sokoban Benchmark: GPT-6.2 Models Clear All Levels(3 posts)→
More from Models
- Timed eval task asks users to deliberately make Claude fail at real work in 12 minutes — Icy-Measurement10 · 2026-11-22
- Tesla FSD 15's real-time view isn't 3D rendering — it's the AI model's own synthetic perception — mark_k · 2026-11-22
- Gemini Black Friday half-price: JP-region trick gets Pro annual plan for $91.54 — huangyun_122 · 2026-11-22
- Microsoft's USB4 RDMA lets mini PCs pool memory to run Qwen3-235B — JFPuget · 2026-11-22
- Google AI Mode now enforces usage limits, users surprised by rate-limit popup — elvano124 · 2026-11-22
- Sol 6.2 now refuses reverse engineering that Sol 6.1 happily did for a month — cardak98 · 2026-11-22