ARC Prize verifies Claude Fable 5.2 at 97.7% on ARC-AGI-3, a new high-score record
Southern-Break5505 · reddit · 2026-11-10
ARC Prize officially verified Anthropic's Claude Fable 5.2, setting a new ARC-AGI-3 high score in the standard harness. Verified results: 97.7% on ARC-AGI-3 at $18.6K with the provider adapter harness (81.2%, $21.9K on the standard harness), 93.8% on ARC-AGI-2 at $0.98/task, and 98.5% on ARC-AGI-1 at $0.36/task. The leaderboard shows Fable 5.2 well ahead of other frontier models at the high-cost end.
Related event: Claude Fable 5.2 Sets New Record on ARC-AGI-3(2 posts)→
More from Models
- Thinking Machines' Inkling 2 Small hits OpenRouter free, 594B MoE with 1M context — soumithchintala · 2026-11-10
- The math behind Anthropic's 5-minute and 1-hour prompt cache options — bigaiguy · 2026-11-10
- Reddit accuses Artificial Analysis of 're-calibrating' benchmarks to bury open-weight models — Rare_Subject2343 · 2026-11-10
- antirez: Mythos uncovered two severe bugs in 20 minutes of testing — antirez · 2026-11-10
- "You get what you RL on": Claude 5.2's robot control generalization sparks talk — chris_j_paxton · 2026-11-10
- Anthropic: Claude Fable/Mythos 5.2 voice suspicion they're being evaluated, muddying safety tests — ChrisGPT · 2026-11-10