GPT-5.6 Sol tops ARC-AGI-3 after OpenAI enables multi-context reasoning
August 5, 2026Posted by Federico Ulfo
GPT-5.6 Sol is now state of the art on ARC-AGI-3 — not through a new checkpoint, but through harness changes. OpenAI allowed the model to reason across multiple context windows, using its canonical compaction implementation, and the score tripled.
OpenAI documented the run in "How two settings tripled our ARC-AGI-3 scores".
Sources:
- OpenAI — How two settings tripled our ARC-AGI-3 scores
- @thsottiaux on X — ARC-AGI-3 SoTA via compaction settings
Related: Sakana AI and UC Berkeley's RHI paper makes the same "the harness now matters as much as the model" point from the research side.