AI & Technology

Anthropic cuts Fable 5.2 token prices 20%, but effort level sets the bill

Written by
Full Name
November 13, 2026
Independent tests by Artificial Analysis put a single Claude Fable 5.2 task at between £2.30 and £20 ($3.10 to $26.54) depending on effort, so the setting chosen by marketing teams, or their software vendors, now matters more to the bill than Anthropic’s lower rate card.

Anthropic’s price cut on its most capable public model took about a day to get complicated. Claude Fable 5.2, released on 9 November alongside a restricted twin, Claude Mythos 5.2, costs about £6 ($8) per million input tokens and £30 ($40) per million output tokens. That is 20% below the $10 and $50 Anthropic charged for Fable 5.1, which launched in September. Fable is now on its third version since Anthropic opened the model line to general use in June.

On 10 November, the independent benchmarking firm Artificial Analysis placed Fable 5.2 first on its Intelligence Index, with a score of 65 at the model’s maximum effort setting. The same testing ranked it the most expensive model on the index to run, at an average of $26.54 per task, with each task taking about 47 minutes. At its lowest setting, Fable 5.2 scored 54 for $3.10 a task in about six minutes. For marketing teams paying for AI by usage, whether directly through Anthropic’s API or through content and analytics tools built on Claude, the per-token price now says less about the final invoice than the effort setting does.

What did Anthropic release, and who can use it?

Anthropic released Claude Fable 5.2 and Claude Mythos 5.2 on 9 November 2026 and describes them as the same underlying model with different safeguards. Fable 5.2 runs classifiers that block high-risk cybersecurity work and most research-biology tasks, which lets Anthropic offer it generally. It is available in the Claude apps, on the Claude Platform, Amazon Bedrock, Google Cloud and Microsoft Foundry under the API identifier claude-fable-5-2.

Mythos 5.2 is limited to organisations approved through Anthropic’s Cyber Verification Program or Life Sciences Verification Program. Claude Enterprise customers can also reach it inside Claude Security, Anthropic’s code-scanning product, without joining either programme.

Anthropic says both models generate output up to 30% faster than the 5.1 versions and improve on coding, reasoning and vision. Its developer documentation lists a 1M-token context window and a maximum output of 128,000 tokens. An optional fast mode, which the company says produces tokens up to 2.5 times faster, costs double the standard rate at $16 and $80 per million tokens. Anthropic also reports about 55% fewer blocked requests than Fable 5.1 in early testing, with any query that is still blocked answered by an older model instead.

The launch post does not position Fable as an everyday tool. Anthropic tells customers to start most coding and knowledge work on Claude Opus 5.5, which it calls its strongest daily driver, and to move up to Fable or Mythos only when a task’s scope or complexity demands it.

Does the price cut make Fable 5.2 cheaper to run?

Fable 5.2 is cheaper per token but not necessarily per job, because the model can spend far more tokens on a task when it is given room to think. Artificial Analysis, which runs its own evaluations rather than relying on vendor figures, found the model used 275 million output tokens to complete its index at maximum effort. That placed it among the most verbose models it has tested and drove the per-task cost to the top of its table.

Coverage of the same data puts Fable 5.1’s equivalent cost at about $12.99 a task. On that comparison, the newer model can cost roughly twice as much per completed task at maximum effort, despite the lower token price.

The picture changes sharply further down the scale. At the high setting, Artificial Analysis scored Fable 5.2 at 62, for $9.08 a task and about 17 minutes. Medium effort scored 56. Opus 5.5 sits at 59 on the same index, and Google’s Gemini 4 Carbon scores 60 at its high setting for $7.76 a task.

Anthropic’s own launch figures point the same way, though on benchmarks it chose and ran itself. The company says Fable 5.2 at high effort matched Opus 5.5’s best score on its FrontierCode 2.0 coding evaluation at roughly half the cost per task. It also claims the model matched Opus 5.5’s top score on one search benchmark at about 40% of the cost, when run at medium effort.

The practical effect for a marketing operations team is that two departments using the same model can face very different bills. A workflow that summarises campaign data or drafts briefs rarely needs a 47-minute reasoning run. Effort is set per request, either by the team itself or by the vendor whose tool sits on top of the model.

What do the claims about black holes and Fermat’s Last Theorem show?

Anthropic’s science results are its own claims and had not been independently confirmed at the time of writing. The company says Fable 5.2 solved two open problems in theoretical physics concerning black holes, with preprints linked from its announcement. It also says a team of 1,000 Fable 5.2 agents formalised the proof of Fermat’s Last Theorem in 30 hours with minimal human intervention.

Formalisation means encoding an existing proof so a computer can check every step. It does not mean discovering a new proof. A footnote Anthropic added to the announcement on 11 November states that the agents began with 60 statements, the relevant definitions, a one-sentence sketch of the classical approach and 47 lemmas that had already been proved.

Anthropic says the UK AI Security Institute, METR and Frontier Design tested the models before release, but its post also says none of them reviewed the announcement or the system card. The company reports that both models voiced suspicion of being evaluated more often than its recent models, which Anthropic itself lists as a limitation of its evaluations.

What can be confirmed is commercial: a lower token price, faster output, wider cloud availability and fewer refusals. The performance case rests on Anthropic’s own tables and, so far, one independent index.

Pro plans and standard Team seats can use Fable within up to half of their weekly usage limits until 16 November, after which Fable use on those plans draws on paid usage credits.

Subscribe to our newsletter

By subscribing you agree to with our Privacy Policy
Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.
Share article

Recommended Reading