Claude Fable 5.2: 20% price cut, up to 30% faster

What Claude Fable 5.2 is and how it relates to Mythos 5.2

Anthropic moved its top Claude model from version 5.1 to 5.2 on November 9, 2026. Fable 5.2 is the same model as the restricted Mythos 5.2 with cybersecurity and biology safeguards added, so any API customer or paid subscriber can use it, and it outscores Opus 5.5 on the FrontierSWE v2 coding benchmark. It shipped the same day on AWS, Google Cloud and Microsoft Azure, and it writes output up to 30% faster than Fable 5.1.

Anthropic also shared what the model did before launch. A team of 1,000 Fable 5.2 agents formalized the proof of Fermat's Last Theorem in 30 hours, a job that took an earlier research version of Claude 11 days. It also solved two open problems in black hole physics and designed three working electronic devices from scratch.

Same model as Mythos 5.2, with cyber and biology safeguards

Fable 5.2 and Mythos 5.2 share one underlying model. Mythos 5.2 goes only to organizations that pass Anthropic's Cyber or Life Sciences Verification Program. Classifiers in Fable 5.2 block requests such as penetration testing and exploit generation, yet it blocks about 55% fewer requests than Fable 5.1.

Anthropic says the safeguards are as robust as before but trigger fewer unnecessary blocks, with about 60% fewer in cybersecurity and 40% fewer in biology. When a request is blocked, Opus 4.8 answers cybersecurity questions and Opus 5 answers biology ones instead.

How Fable 5.2 compares with Opus 5.5 on coding benchmarks

FrontierSWE v2 score (%)

From Anthropic's comparison table. Claude scores were measured with safeguards on.

0204060

69.9Fable 5.2

65.5GPT-6 Astra

62.3Opus 5.5

56.3Fable 5.1

Anthropic lists FrontierSWE v2 under agentic coding in its comparison table. Fable 5.2 scores 69.9%, ahead of GPT-6 Astra at 65.5% and Opus 5.5 at 62.3%. Fable 5.1 scores 56.3%.

Results are closer on FrontierCode 2.0, which checks whether maintainers of open-source repositories would merge a model's code changes. With effort (the setting that controls how deeply the model reasons) set to high, Fable 5.2 scored 45.3% against Opus 5.5's best score of 45.5%, at roughly half the cost per task. Fable 5.2 also trails Gemini 4 Carbon on AutomationBench, 51.4% to 53.9%.

Claude Fable 5.2 API pricing and fast mode

API usage is billed per token, the small chunks of text a model reads and writes. Fable 5.2 charges $8 for every million tokens it reads and $40 for every million it writes. Fable 5.1 cost $10 input and $50 output, so both rates fell 20%.

Cache reads, which bill repeated context at a discount, cost $0.20, or 0.025 times the input price, the same multiplier as on Fable 5.1. Batch requests are half price.

Fast mode doubles the price for up to 2.5x faster output

Fast mode is an optional setting that trades a higher rate for speed. Fast mode costs $16 input and $80 output per million tokens and works in Claude Code, the Claude Platform, Google Cloud and Microsoft Foundry. Token generation runs up to 2.5 times faster than the standard rate allows.

Fable 5.2 still costs twice as much per token as Opus 5.5

Opus 5.5 lists at $4 input and $20 output, half the Fable 5.2 rate. The gap narrowed from 2.5x on Fable 5.1, and on FrontierCode 2.0, Fable 5.2 at high effort cost about half as much per task as Opus 5.5.

Per-token price is only part of the cost, because the number of tokens a task uses matters too. Anthropic still recommends that most users start on Opus 5.5. For its pricing and benchmarks, see Claude Opus 5.5 pricing, benchmarks, API changes.

Launch offers for Pro, Team, Max and Ultra plans

To mark the launch, Anthropic gave paid subscribers extra Fable usage that covers Fable 5.2, Fable 5.1 and Fable 5. Pro and Team standard seats can spend up to 50% of their weekly limits on Fable until November 16 at 11:59 pm PT, plus one Fable reset if they were paying before launch. The billing cutoff for the reset is November 9, 2026 at 10:00 am PT.

A reset gives back the usage you have already spent on Fable models. It doesn't raise your limits.

Max and Ultra plans get one to three Fable resets

Higher plans get resets rather than a week of included usage. Max, Ultra and premium seats get one reset under six months of paid subscription, two at six to 12 months and three beyond 12 months, all expiring March 8, 2027. Subscription length is counted from your first paid bill, and unused resets lapse at the deadline.

Pro and Team go back to usage credits after November 16

Once the launch week ends, Pro and Team standard seats can use Fable only with usage credits turned on. That is how Fable worked for these seats before launch. Any unused part of the 50% allowance doesn't carry over. For how Fable returned to paid plans earlier this year, see Is Claude Fable 5 Back? Yes, With New Safeguards.

Four API changes to check before you migrate

Swapping the model ID from claude-fable-5-1 is not always enough. Anthropic lists four changes that can break existing code: three make requests fail with an error, and the fourth loses reasoning when a conversation moves to another model.

If Claude Code, claude.ai or the Claude Agent SDK builds your requests, nothing changes, so only code that builds its own API requests needs checking. Effort also defaults to high when the API request doesn't set it, while Claude apps and Claude Code default to medium.

Edited conversation history now fails on every account

Thinking blocks hold the reasoning a model produced before it answered. The API's prefix check confirms that nothing before a thinking block, such as the system prompt, the tool list or an earlier message, has changed since the model wrote it.

Fable 5.1 enforced the prefix check only on accounts created on or after August 31, 2026, but Fable 5.2 enforces it on every account. A request that fails the check is rejected, so the fix is to keep conversation history append-only.

Update computer use and advisor tool settings

Computer use lets Claude operate a desktop through tool calls. The old computer_20251124 tool now errors everywhere except Amazon Bedrock, so integrations need computer_toolset_20260801.

The advisor tool, still in beta, lets one model consult another, at least as capable, on hard problems. With a Fable 5.2 executor, the advisor must be Fable 5.2 or Mythos 5.2, and a request that points it at Fable 5.1 is rejected.

Start on Opus 5.5 and move up to Fable 5.2 only when needed

Fable 5.2 beats its predecessor on speed, price and coding scores, but the per-token gap to Opus 5.5 is still 2x, and the included Pro and Team usage ends on November 16. The cheapest path is to keep everyday work on Opus 5.5 and use the launch week to find the tasks that actually need Fable 5.2.

On the API, run through the "Four changes when moving API code from Fable 5.1" table before you change the model ID. Another option Anthropic describes is to keep Haiku, Sonnet or Opus as the main model and call Fable 5.2 only as an advisor on the hardest problems.