ARC Prize publishes evaluation results of Claude Fable 5.2 on ARC-AGI

ARC Prize published Anthropic's Claude Fable 5.2's results on the ARC-AGI series of benchmarks: under the ARC-AGI-3 standard harness it scored 81.2% at a cost of $21.9K, and under the provider adapter harness 97.7% at $18.6K; ARC-AGI-2 was 93.8% at $0.98 per task, and ARC-AGI-1 was 98.5% at $0.36 per task. ARC Prize says Fable 5.2 set a new high score on ARC-AGI-3 under the standard harness, and its 93.8% on ARC-AGI-2 is the third highest among models seen so far, behind only GPT-6 Astra's 95.0% and GPT-6.1 Sol's 94.2%. At the max reasoning tier with the standard harness, 28 of the 55 semi-private games hit the 12-hour runtime limit, scoring 65.7%, lower than the 81.2% at the xhigh tier; the released Fable 5.2's tuning at the medium and xhigh tiers differs from the previously evaluated checkpoint, while the low, high, and max tiers are unchanged.