ARC Prize publishes Claude Fable 5.2 ARC-AGI results, max effort trails xhigh

1. 2:13 PM
   ARC Prize@arcprizeOfficial

   ARC Prize publishes full Claude Fable 5.2 benchmark results

   AIARC Prize has published full benchmark results for Claude Fable 5.2 on its leaderboard, along with a reproducible evaluation repository and its testing policy. The post links to the complete results page for the model.

2. 2:13 PM
   ARC Prize@arcprizeOfficial

   ARC Prize reports Fable 5.2 scores 93.8% on ARC-AGI-2

   AIARC Prize says Anthropic's Fable 5.2 scored 93.8% on ARC-AGI-2, the third-highest model score it has recorded. GPT-6 Astra scored 95.0% and GPT-6.1 Sol scored 94.2%, ranking ahead of Fable 5.2.

   
3. 2:13 PM
   ARC Prize@arcprizeOfficial

   ARC Prize says Fable 5.2 max scored 65.7% on ARC-AGI-3, below xhigh's 81.2%

   AIARC Prize reports that Fable 5.2 scored 65.7% on ARC-AGI-3 at max reasoning effort, versus 81.2% at xhigh. It says 28 of 55 semi-private games hit the 12-hour runtime limit, and it expects max would have scored higher without that cap. The released Fable 5.2 also uses different tuning at medium and xhigh than the earlier checkpoint ARC Prize evaluated, while low, high, and max are unchanged.

4. 2:13 PM
   ARC Prize@arcprizeOfficial

   ARC Prize reports Claude Fable 5.2 scores on ARC-AGI benchmarks

   AIARC Prize reports verified ARC-AGI results for Claude Fable 5.2 from Anthropic. The post lists 97.7% on ARC-AGI-3 at $18.6K with the provider adapter harness and 81.2% at $21.9K with the standard harness. It also reports 93.8% at $0.98 per task on ARC-AGI-2 and 98.5% at $0.36 per task on ARC-AGI-1, and says Fable 5.2 sets a new high score on ARC-AGI-3 in the standard harness.

   

The trend compares only continuously observed participants, so its scope may be smaller than the current heat total. Hover or tap for hourly heat; use arrow keys with a keyboard.