Claude Fable 5.2 cuts token prices 20% from Fable 5.1

Anthropic released Claude Fable 5.2 and Claude Mythos 5.2 on November 9, 2026, and cut their price to $8 per million input tokens and $40 per million output tokens. That is 20% less than the 5.1 models, CNBC reports. Tokens are the small chunks of text a model reads and writes, and they are how the API bills. For teams that already pay for Fable, the same 1-million-token memory now costs less. But the new model also comes with rules that can break existing code.

The two names cover one model. Anthropic's announcement says Fable is the public version, with extra safeguards for cybersecurity and biology. Mythos is the same model for vetted organizations in Anthropic's Cyber and Life Sciences Verification Programs.

What Anthropic released

Anthropic says Fable 5.2 writes its answers up to 30% faster than Fable 5.1. Its safeguards also block about 55% fewer requests than before. On FrontierSWE v2, a coding test, Anthropic lists 69.9% for Fable 5.2, 56.3% for Fable 5.1 and 62.3% for Claude Opus 5.5. These are Anthropic's own numbers.

Anthropic still does not pitch Fable as the everyday model. It names Opus 5.5 as the daily driver for most coding and everyday knowledge work, and keeps Fable 5.2 for the hardest jobs. CNBC notes these are the first frontier models Anthropic has shipped since its CEO, Dario Amodei, called in September for AI work to slow down. "We called for pacing, not pausing," Dianne Penn, Anthropic's head of product management for research, told CNBC. The launch also lands just before Anthropic's planned stock market debut.

Fable 5.1 vs Fable 5.2, rate by rate

The table below comes from Anthropic's pricing page, its fast mode page and the two models' overview pages. Prices are US dollars per million tokens. A cache write stores a long prompt so later calls can reuse it, and a cache hit is that cheaper reuse.

	Fable 5.1	Fable 5.2
Input	$10	$8
Output	$50	$40
Cache write, 5 minutes	$12.50	$10
Cache write, 1 hour	$20	$16
Cache hit	$0.25	$0.20
Batch API	50% off input and output	50% off input and output
Fast mode	Not listed	$16 input, $80 output
Context window	1M tokens	1M tokens
Maximum output	128K tokens	128K tokens
Knowledge cutoff	June 2026	June 2026
Released	September 1, 2026	November 9, 2026
Retired no sooner than	September 1, 2027	November 9, 2027

Every rate drops by the same 20%. So a cache hit still costs 2.5% of the input price, and the case for caching long prompts stays the same. Mythos 5.2 has the same rates as Fable 5.2. Fast mode, which Anthropic says makes output up to 2.5 times quicker, doubles the bill. The fast mode page also says it does not work with the Batch API.

Fable 5.2's model page lists three limits that matter for code. A request that forces a tool call with tool_choice set to any or tool returns a 400 error. The temperature, top_p and top_k settings must be left out. The model also needs 30-day data retention, so it is not offered under zero data retention unless Anthropic approves it.

Where to use it and how to switch

Anthropic says Fable 5.2 runs on every platform it sells through. The IDs and conditions below come from Anthropic's model pages, the Claude Code changelog, the GitHub Changelog and Anthropic's support page on launch offers.

Where	How to get Fable 5.2	Condition
Claude API	claude-fable-5-2	Mythos uses claude-mythos-5-2 after verification
Amazon Bedrock	anthropic.claude-fable-5-2	None listed
Google Cloud, Microsoft Foundry, Claude Platform on AWS	claude-fable-5-2	Foundry fast mode is claude-fable-5-2-fast
Claude Code	Default Fable model from version 2.1.321	/model claude-fable-5-1 switches back
GitHub Copilot	Model picker	Pro+, Max, Business and Enterprise; off by default on Business and Enterprise
Claude apps, Pro and standard Team	Pick Fable in the model picker	Included until November 16, 11:59 pm PT, up to 50% of weekly limits
Claude apps, Max, Ultra, premium Team	Pick Fable in the model picker	One to three Fable limit resets, usable until March 8, 2027

GitHub says Copilot keeps Fable 5.2 data by default so Anthropic can run safety checks, and that the data is not used for training. Some enterprises can get zero retention until the end of the year. After the included week, Pro and standard Team seats pay for Fable with usage credits, Anthropic's support page says.

What this means for developers

Moving from Fable 5.1 to 5.2 is a cheaper model with the same context. But it is not a one-line swap for every codebase. Before you change the model ID, check a few things:

• Search your code for sampling settings. Any call that sets temperature, top_p or top_k must drop them for Fable 5.2. Test agent code that forces a tool call, since that now returns a 400 error.
• Check your data terms. If your company relies on zero data retention, Fable 5.2 is off the table unless Anthropic signs off. Copilot Business and Enterprise admins also have to turn the model on first.
• Redo the cost math. On a job with 2 million input and 500,000 output tokens, Fable 5.1's list price is $45 and Fable 5.2's is $36. Fast mode on 5.2 would cost $72 for the same job.
• Compare it with Opus 5.5 first. Opus 5.5 lists at $4 and $20, half of Fable 5.2. Anthropic also says Fable 5.2 at high effort scored 45.3% on FrontierCode 2.0, about level with Opus 5.5, at roughly half the cost per task. Run your own tasks on both before you move everything.
• Pin the version in Claude Code. Saved Fable 5.1 choices now move to the fable alias, which points at 5.2. Use /model claude-fable-5-1 if you need to compare old and new output side by side.

Pro subscribers have until November 16 to try Fable without spending credits. Teams that lean on caching gain twice: Fable 5.1 already made cache reads cheaper in September, and 5.2 trims them again, from $0.25 to $0.20.