GPT-6.2 Astra scores 100% on ARC-AGI-3, first model to max the benchmark
patience_cave · x · 2026-11-18
- ARC Prize published verified ARC-AGI results for OpenAI's GPT-6.2 Astra: a perfect 100.0% on ARC-AGI-3 (SOTA), the first model ever to max the benchmark, costing $4.1K via provider adapter; standard mode scores 70.3% at $24.2K.
- It also leads ARC-AGI-2 at 95.8% (SOTA, $0.50/task) and hits 98.5% on ARC-AGI-1 at $0.17/task.
- The leaderboard chart shows Claude Fable 5.2 and Gemini 4 variants still trailing well behind on ARC-AGI-3.
More from Models
- GPT-6.2 Astra hits near-parity with Fable 5.2 at far lower cost, reigniting frontier race debate — daniel_mac8 · 2026-11-18
- GPT-6.2 fails to beat GPT-6 on Coding Agent Index, trailing Claude and Gemini — realsohamparekh · 2026-11-18
- Teknium flags inconsistent OpenAI demos: was GPT-6 quietly nerfed before GPT-6.2? — Teknium · 2026-11-18
- GPT-6.2 Astra beats Opus 5.5 across agent benchmarks at 3-6x lower cost — VraserX · 2026-11-18
- Side-by-side: GPT-6.2 Astra builds noticeably richer scenes than GPT-6 Astra — Angaisb_ · 2026-11-18
- 3 weeks with OpenAI Astra 5.2: shines at delegating under messy stacked context — round · 2026-11-18