Claude Fable 5.2 Launches With Cheaper Tokens and Top AI Score

Anthropic says its prerelease testers used the new models for tasks that would otherwise have been out of reach, including completing software migrations, designing three working electronic devices from scratch and solving two open problems in black hole physics. The company also says the models formalized the proof of Fermat's Last Theorem in 30 hours, a claim reported by The Verge.

On benchmarks, Anthropic reports that Claude Fable 5.2 sits above Claude Opus 5.5 in almost every capability tested. On FrontierCode 2.0, which checks whether a model's code changes would be merged by open-source maintainers, Fable 5.2 reached 49.1% at Xhigh effort, the highest score of any released model Anthropic tested, against 42.8% for GPT-6 Astra. On Humanity's Last Exam, the source lists GPT-6 Astra at 71.0% and Gemini 4 Carbon at 72.5% at Max effort.

Artificial Analysis paints a more mixed picture on speed and cost. Its independent testing measured Claude Fable 5.2 at 215.95 seconds to first answer token, 98 tokens per second output speed and 46.4 minutes per Intelligence Index task, which it describes as notably slow compared with a median of 12.6 minutes. It also flags the model as very verbose and expensive relative to similar models, with an average cost of $26.54 per task on the Intelligence Index.

For anyone using Claude in Tunisia, the practical change is price and speed rather than access. Claude Fable 5.2 is available today through Anthropic, and the 20% cut to token pricing plus the optional fast mode should lower the cost of heavy coding or research sessions. There is no announced Tunisian pricing or local availability, so users here would pay the same dollar rates through Anthropic's API or Claude apps, roughly 23 TND per million input tokens and 116 TND per million output tokens at 1 USD ≈ 2.9 TND.

Anthropic recommends that most users start with Claude Opus 5.5 for everyday coding and knowledge work, then move up to Fable or Mythos only for ambitious, open-ended projects. If your work involves cybersecurity or biomedical research, Mythos 5.2 requires applying to Anthropic's verification programs, and approval is not guaranteed.