How enabling two settings tripled our scores on the ARC-AGI-3 benchmark

OpenAI found that enabling two simple API settings on GPT-5.6 tripled its performance on the ARC-AGI-3 benchmark. The settings, reasoning retention and compaction, improved both scores and efficiency.

OpenAI

Claude Sonnet 5 Expected Imminently, With Major Upgrades

Anthropic is expected to release Claude Sonnet 5 soon, with major upgrades in coding, reasoning, and safety. The new model will be free and outperform GPT-4o in key areas.

Anthropic OpenAI