Rankings are cheap to read and expensive to act on. Claude Fable 5.2 (Max) tops Artificial Analysis at 65, against a median of 40, but it costs $26.54 per Intelligence Index task at $8 in and $40 out per million tokens. For a build pipeline, that makes it a model for hard calls, not a default.
What I’d do: keep a cheaper model as the default in any agent loop, and send Fable only the work where a wrong answer costs more than the tokens. Its 46.4 minutes per task against a median of 12.6 is what would break a build pipeline first. Cap output tokens per run and log cost per task, not just quality.
The story — Artificial Analysis lists Claude Fable 5.2 (Max), a proprietary reasoning model from Anthropic released November 2026, at 65 on its Intelligence Index, ranked #1 of 74 models against a median of 40. It takes text and image input, outputs text, and has a 1M-token context window. Pricing is $8.00 per million input tokens and $40.00 per million output tokens. The page also reports 275M output tokens against a median of 72.8M. (Source)