Frontier Log

Benchmark · Superseded

ARC-AGI-1

020406080100Jan 2026Apr 2026Jul 2026Oct 202686.2%GPT-5.293.7%GPT-5.495.0%GPT-5.598.5%GPT-6 Astra020406080100Jan 2026Apr 2026Jul 2026Oct 202693.7%GPT-5.498.5%GPT-6 Astra

Results

ModelScoreSourceHuman-verified
GPT-6 Astra3 Sep 2026 · OpenAI98.5%GPT-6 Astra: A new generation of intelligenceRow label 'ARC-AGI-1' (no 'Verified' printed). The fetched page text was truncated, so footnotes without a % sign are not visible in the extract.Yes
GPT-5.523 Apr 2026 · OpenAI95.0%Introducing GPT-5.5Row label 'ARC-AGI-1 (Verified)'. The fetched page text was truncated, so footnotes without a % sign are not visible in the extract.Yes
GPT-5.45 Mar 2026 · OpenAI93.7%Introducing GPT-5.4Row label 'ARC-AGI-1 (Verified)'. The fetched page text was truncated, so footnotes without a % sign are not visible in the extract.Yes
GPT-5.211 Dec 2025 · OpenAI86.2%Introducing GPT-5.2Row label 'ARC-AGI-1 (Verified)'. Repeated in the detailed abstract-reasoning table (row 'ARC-AGI-1 (Verified)', column 'GPT-5.2 Thinking'). The fetched page text was truncated, so footnotes without a % sign are not visible in the extract.Yes

Benchmark

arcprize.orgHigher is better

François CholletIntroduced 5 Nov 2019

All benchmarks