ChatGPT5.5 with a tool harness cracks Sokoban: 68 tool calls in 3 minutes
tak3sh8 · x · 2026-09-26
The author stress-tested ChatGPT5.5 on Sokoban puzzles: a bare first attempt (no setup, 10 minutes in) was already encouraging. After adding tools to the agent harness, one run made 68 tool calls in 3m 5s and performed markedly better — a demonstration of how bolting tools onto an agent harness amplifies a model's puzzle-solving ability.
More from coding & agent
- Stealth model Space Bunny rebuilds entire site from a 40s screen recording — PrajwalTomar_ · 2026-09-26
- AI veteran: Agentic AI is the control harness, not a separate business opportunity — nicolaiwadstrom · 2026-09-26
- McKinsey Deck: a Claude Skill that builds editable consulting-style PPT slides — baabullah · 2026-09-26
- KoboldCpp ships built-in Agent harness; author warns of phishing site koboldcpp.com — HadesThrowaway · 2026-09-26
- Nace launches Drex, a sub-6B decision model outputting option probabilities instead of prose — rohanpaul_ai · 2026-09-26
- The team was already running LLM cross-checks on reports — the real gap was that it all lived in one person's account — beglen · 2026-09-26