Claude Fable 5.2 Tops Artificial Analysis Index With Score of 65, but Costs More per Task

The result, published on the day of the model's release, gives outside support to Anthropic's own benchmark claims a week before the company's planned Nasdaq listing. Most of the figures Anthropic published at launch were measured by Anthropic.

Artificial Analysis ranks Fable 5.2 first of the 74 models in its intelligence comparison, against a median of 40 for comparable models. According to The Decoder's analysis of the results, the nearest rival is Google's Gemini 4 Carbon at its high setting with 60 points; Google has not yet released that model. Anthropic's Claude Opus 5.5 scored 59 at its maximum setting, and the previous Fable 5.1 trailed the new model by 15 points.

The lead holds at lower settings. Artificial Analysis lists 62 points for Fable 5.2 at both its high and extra-high effort levels, 56 at medium and 54 at low. The Decoder said Fable 5.2 led five of the nine tests that make up the index, but trailed Opus 5.5 and Sonnet 5.5 on the AA-Briefcase test and lagged Kimi K3.1 Pro and Gemini 4 Carbon on Terminal-Bench 4.0.

The top score is expensive to reach. Artificial Analysis puts the cost of the maximum setting at $26.54 per index task, the highest of the 73 models it priced. The Decoder compared that with about $10 for Opus 5.5 and $12.99 for Fable 5.1. Fable 5.2 also took an average of 44.1 minutes per task, against a median of 12.9 minutes, and its output speed was measured at 103 tokens a second.

List prices are lower than for earlier Fable and Mythos models. Anthropic charges $8 per million input tokens and $40 per million output tokens, down from $10 and $50. Anthropic has said it still recommends Opus 5.5 for everyday use.

Fable 5.2 is available to all Anthropic customers. The more capable Mythos 5.2 is limited to vetted organisations in life sciences and cybersecurity and does not appear in the index rankings. Anthropic is expected to file pricing details for its IPO within days.