Fable 5.2 tops Epoch and Remote Labor Index, but not every test

In the days after Anthropic released Claude Fable 5.2, independent evaluators put it at or near the top of most leaderboards. Epoch AI scored it 174 on its Epoch Capabilities Index at launch, a record above GPT-6 Astra's 169 and above what Epoch's pre-Astra trend for reasoning models predicted; its model page now lists 173, still first of 240 tracked models.

The biggest jump came on the Remote Labor Index, built by the Center for AI Safety and Scale AI from real Upwork projects judged against a paid professional's work. Fable 5.2 fully automates 41.3% of them, per CAIS and Scale's leaderboard, up from 15.8% for Fable 5 in June and 2.5% for the best agent when the benchmark launched a year ago. CAIS says the rate is doubling about every three months; Benjamin Todd projects the index will be saturated by April 2027.

Other results were more mixed. P-Zero Research found it the best model yet at predicting experiment results, at 0.36, and it took second place behind GPT-6 Astra on RuneScapeBench. In Dwarkesh Patel's Blood on the Clocktower tournament, a social deception game, it beat Opus 5.5 and Fable 5.1 but still trailed Astra. On Artificial Analysis's run of Harvey's LAB legal benchmark it passed 10.8% of tasks, below Fable 5.1 (14.4%) and Sonnet 5.5 (11.7%).

Anthropic's own headline claim is scientific: its launch post says Fable 5.2 solved two open problems in black-hole physics, working largely autonomously on one, a calculation laid out in 2003 that confirmed black holes form abruptly in very small universes too. Both results are in preprints and have not yet been checked by outside physicists.