6 on the ARC-AGI-3 benchmark. The two settings, described as "retaining reasoning" and "enabling compaction," boosted both scores and efficiency. The changes resulted in a significant performance improvement on the ARC-AGI-3 benchmark.
OpenAI Blog · Models
How enabling two settings tripled our scores on the ARC-AGI-3 benchmark
Wednesday, July 29, 2026
Read Original Source
← Back to AI News