FrontierJuly 29, 2026via OpenAI Blog
How enabling two settings tripled our scores on the ARC-AGI-3 benchmark
Why it matters
OpenAI disclosed a significant capability unlock on a major reasoning benchmark by toggling reasoning retention and compaction settings — practical tuning that practitioners may need to adopt to match published performance claims.
Key signals
- GPT-5.6 scores tripled on ARC-AGI-3 with two API settings
- Settings: reasoning retention and compaction enabled
- Benchmark: ARC-AGI-3 (abstract reasoning task)
- Source: OpenAI official blog post
- Date: July 29, 2026
- Implication: published benchmarks may depend on non-default configurations
The hook
Two API toggles tripled GPT-5.6's score on ARC-AGI-3. Here's what changed.
How two API settings improved GPT-5.6 performance on ARC-AGI-3, boosting scores and efficiency by retaining reasoning and enabling compaction.