AnalysisAI ModelsJuly 29, 2026
How enabling two settings tripled our scores on the ARC-AGI-3 benchmark
Two API settings on GPT-5.6 — retaining reasoning and enabling compaction — tripled its ARC-AGI-3 scores and improved efficiency, per OpenAI.
3 sources
OpenAI by email
Get an email when OpenAI ships something
More stories today
- Y Combinator open-sources QM, its multiplayer agent harness
- Prompt idea: Make up a random word and have ChatGPT draw it
- Developer claims Claude Code now handles 95% of his work
- ChatGPT users complain about heavily caveated answers
- Semantica provides open-source enterprise intelligence layer for AI agents