GPT-6.2 Astra scores first-ever 100% on ARC-AGI-3 via provider adapter
mark_k · x · 2026-11-18
Per markk with an ARC Prize-verified leaderboard, OpenAI's GPT-6.2 Astra became the first model to score a perfect 100% on ARC-AGI-3, plus 95.8% on ARC-AGI-2 and 98.5% on ARC-AGI-1. GPT-6.2 Sol hit 98.2% on ARC-AGI-3 at 64% lower cost. The caveat: the perfect score required a provider adapter; Astra's standard configuration scored 70.3%, so it's an extraordinary milestone but not proof of AGI. Verified runs cost on the order of tens of thousands of dollars.
More from Models
- Microsoft ships MAI Code 2 and Code 2 Flash coding models into GitHub Copilot — marlene_zw · 2026-11-18
- OpenAI's GPT-6.2 Astra jumps 22 points on ExploitBench, ships tighter cyber safeguards — JiaweiLiu_ · 2026-11-18
- Matt Shumer: this model release needed 'a lot more iteration' — mattshumer_ · 2026-11-18
- GPT-6.2 Astra early review: best-in-class 3D, but frontend styling needs babysitting — flavioAd · 2026-11-18
- OpenAI launches GPT-6.2 Astra and Sol; early user recreates cross-section book visuals — craigsdennis · 2026-11-18
- Sebastien Bubeck: GPT-6.2 as a science partner is 'pretty magical' — SebastienBubeck · 2026-11-18