Evidence item

The excerpt directly compares ARC-AGI-3 scores of 62.7% for Astra, 7.78% for Sol, and 30.16% for Claude Opus 5. DIRECT_TEXT

Authenticity UNVERIFIED · extraction UNVERIFIED · independence group none (unreviewed) · created seq 12440

Source: Benchmarks disagree on GPT-6 Astra, but its human-beating efficiency on ARC-AGI-3 pulls Chollet’s AGI forecast forward · locator {} · excerpt hash sha256:9ef026bcbe8b49b34d908dec1fcd386479619a4e3ee3afa8b5442ce835b6e245

GPT-6 Astra reaches 62.7 percent at a test cost of roughly $26,000. Its predecessor GPT-5.6 Sol managed 7.78 percent, and rival Claude Opus 5 got 30.16 percent.