Claude Fable 5.2 launched on November 9, 2026 together with Mythos 5.2. They are the same model: Fable ships with strong cybersecurity and biology safeguards and is open to everyone, while Mythos, with part of those restrictions removed, goes to vetted organizations. At Anthropic, Fable 5.2 costs $8 per million input tokens and $40 per million output tokens, and fast mode costs twice that.
Everything about the model below comes from Anthropic's announcement, the model page, the API and Claude Code documentation, and the subscriber help articles. Speed, benchmark scores and block rates are Anthropic's own figures, and we did not reproduce them. We re-read Anthropic's prices, the model status and the offer terms on November 17, and everything else in the article was checked on November 13.
What is new and when it launched
By Anthropic's documentation, Fable 5.2 is built for demanding reasoning and long-horizon agentic work: sessions that run for hours, multistep research, and analysis carried through to a finished document, spreadsheet or deck. The announcement claims:
- output up to 30% faster than Fable 5.1 and Mythos 5.1, at a lower price;
- an optional fast mode that speeds up token generation by up to 2.5x;
- on FrontierCode 2.0, which measures whether open-source maintainers would merge a model's changes, Fable 5.2 at high effort scores 45.3% against 45.5% for Opus 5.5's best result, for roughly half the cost per task by Anthropic's estimate. At xhigh it reaches 49.1%, the highest of any released model Anthropic tested. GPT-6 Astra's best score in the same comparison is 42.8%;
- about 55% fewer blocks than Fable 5.1 (about 60% fewer in cybersecurity and 40% fewer in biology) in early testing across claude.ai, Claude Code and Cowork.
The safety classifiers remain. Fable 5.2 blocks high-risk cyber work, including penetration testing and exploit generation, and most research biology. In the apps a blocked query gets an answer from Opus 4.8 (cyber) or Opus 5 (biology). In the API a refusal arrives as a normal HTTP 200 response with stop_reason: "refusal", and a retry on another model is configured with the fallbacks parameter. It is in beta, and the parameter alone is not enough: you need an anthropic-beta header. Use server-side-fallback-2026-07-01, which supports both fallbacks: "default" and an explicit list of models, while server-side-fallback-2026-06-01 accepts only the explicit list. The permitted fallback targets for Fable 5.2 are Opus 4.8 (claude-opus-4-8) and Opus 5 (claude-opus-5).
| Parameter | Value in Anthropic's documentation |
|---|---|
| API ID | claude-fable-5-2 |
| Context window | 1M tokens, shared by input, thinking and the response |
| Maximum output | 128K tokens per request, inside the same 1M window |
| Thinking | adaptive, always on |
| Default effort | high in the API, medium in the Claude apps and Claude Code |
| Knowledge cutoff | June 2026 |
| Status | active, retirement not sooner than November 9, 2027 |
| Platforms | Claude API, Amazon Bedrock, Claude Platform on AWS, Google Cloud, Microsoft Foundry |
Mythos 5.2: what it is and who can use it
Mythos 5.2 has the same capabilities as Fable 5.2 and shares its specifications and price ($8 and $40), minus part of the cyber and biology safeguards. Anthropic gives it to organizations that pass verification through its Cyber and Life Sciences Verification Programs. Claude Enterprise customers can use Claude Security, which runs on Mythos 5.2, even if they are not in the Cyber Verification Program. Fast mode for Mythos is open only to members of those programs. The model ID in the status table is claude-mythos-5-2.
To get Mythos, apply to the program that covers your use case or contact your Anthropic account team.
What Fable 5.2 costs
Prices come from Anthropic's pricing page, captured on November 17, 2026, in dollars per 1M tokens.
| Model | Input | Output | Cache read |
|---|---|---|---|
Fable 5.2 (claude-fable-5-2) | $8 | $40 | $0.20 |
Fable 5.1 (claude-fable-5-1) | $10 | $50 | $0.25 |
Opus 5.5 (claude-opus-5-5) | $4 | $20 | $0.20 |
Sonnet 5.5 (claude-sonnet-5-5) | $2 | $10 | $0.10 |
Fable 5.2 is cheaper than Fable 5.1 ($8 and $40 against $10 and $50), and it is exactly 2x Opus 5.5 on both input and output ($8 against $4 and $40 against $20). Cache reads cost the same on Fable 5.2 and Opus 5.5, $0.20.
Anthropic also lists, for Fable 5.2, a 5-minute cache write at $10, a 1-hour write at $16, and a Batch API at half price ($4 and $20). For models 4.7 and newer Anthropic warns about a newer tokenizer that produces about 30% more tokens for the same text, so look at the cost of a whole task: the per-million price alone will not show it.
Fast mode: what it gives and what it costs
Fast mode runs the same model in a configuration with faster output. By the documentation, intelligence and capabilities do not change, the up-to-2.5x gain applies to output tokens per second rather than time to first token, and it is most visible when streaming. It is available on Fable 5.2, Mythos 5.2 and Opus 5.5, and on Opus 5.5 it costs $8 and $40.
| Fable 5.2 mode | Input, $ per 1M | Output, $ per 1M |
|---|---|---|
| Standard | 8 | 40 |
| Fast | 16 | 80 |
| Batch (not combinable with fast) | 4 | 20 |
Fast costs twice the standard rate on both input and output. In the Messages API you turn it on with speed: "fast", and usage.speed in the response shows which speed was applied. The mode has its own rate limits: exceeding them returns a 429 with a retry-after header, and the SDK retries such requests up to two times by default. If you would rather fall back to standard speed than wait, resend the request without speed, but the prompt cache will miss: requests at different speeds do not share cached prefixes. Fast mode cannot be combined with the Batch API or with a Priority Tier commitment. In Claude Code the /fast command toggles it, and on a subscription it is billed through usage credits only, outside the plan limits. On Microsoft Foundry hosted on Anthropic it has its own ID, claude-fable-5-2-fast: you send the request to a deployment of that model and leave speed out. For Foundry hosted on Azure the availability table lists fast mode only for Opus 5.5.
If you go through a third-party LLM gateway, Claude Code's documentation describes how the fast mode availability check can fail behind a gateway, and names the variables CLAUDE_CODE_SKIP_FAST_MODE_ORG_CHECK and CLAUDE_CODE_SKIP_FAST_MODE_NETWORK_ERRORS for that case. We have not tried this setup.
Claude subscriptions: limits and the launch offer
Fable 5.2 limits depend on your Claude subscription plan. Here is a short summary of Anthropic's help article on the offer and of Claude Fable models on your plan. The offer started on November 9, 2026 with the launch and covered every Fable model: 5.2, 5.1 and 5.
| Plan | What the offer gave | Status on November 17 |
|---|---|---|
| Pro and standard Team seats | a week of Fable: all Fable models included in the plan from November 9 to 16, 2026, up to 50% of weekly limits, plus one limit reset | the week ended on November 16, 11:59 pm PT, and the reset expired then |
| Max, Ultra, premium Team seats and premium seats on legacy seat-based Enterprise (HIPAA-ready Enterprise excluded) | one, two or three Fable limit resets by subscription length | unused resets expire March 8, 2027, 11:59 pm PT, or sooner if you cancel or downgrade |
| Free, standard seats on legacy seat-based Enterprise, self-serve Enterprise, usage-based Enterprise | not covered | none |
The number of resets depends on how long the paid subscription has run: under six months gives one reset, six to twelve months gives two, more than a year gives three. Length is counted from the first paid bill, and on Team and Enterprise from the organization's first paid bill, so every premium seat in one organization gets the same number. Resets went only to accounts whose paid billing began before 10:00 am PT on November 9, 2026. For Team and Enterprise seats the organization's billing had to be in place by then, and the seat holder must also have joined the organization before then. If the subscription had started earlier but was still in a free trial, the reset arrived once it converted to paid. Self-serve Team trials count as paid in the help article. Resets are not available on HIPAA-ready Enterprise plans. A subscription started after the cutoff did not get a reset. Resets cannot be transferred or exchanged for money, and they can be removed for fraud, abuse or a violation of the terms.
The Fable week on Pro and standard Team seats ran separately from the reset: the help article ties the billing cutoff to the reset and names no such condition for the week itself. The week ended at 11:59 pm PT on November 16, which is about 07:59 UTC on November 17 (our conversion, with PT at UTC-8 after the November 1 clock change).
A reset returns what you spent on Fable in the weekly limit and refills the five-hour window completely if your account has one. It does not raise the limits. You use it under Settings, Usage, Limit resets, and a used reset cannot be undone. If you cancel or downgrade before using it, the reset is gone even though March 8, 2027 has not arrived.
What applies on November 17, by the pricing page and Anthropic's help article. On Free, Fable is not available. On Pro and standard Team seats it runs on usage credits only (the pricing page says "Fable, paid with usage credits"). On Max, Ultra and premium seats Fable is included up to 50% of the weekly limit, then usage credits or another model. If you are weighing a subscription against pay-per-token access, Claude Pro and Max or the API covers the trade-off, and Anthropic's help article has the current Fable terms.
How Fable 5.2 differs from Opus 5.5 and Sonnet 5.5
The comparison from Anthropic's model page (price per 1M tokens, input and output):
| Parameter | Fable 5.2 | Opus 5.5 | Sonnet 5.5 |
|---|---|---|---|
| Price | $8 and $40 | $4 and $20 | $2 and $10 |
| Context window | 1M | 1M | 1M |
| Maximum output | 128K | 128K | 128K |
| Latency | moderate | moderate | fast |
| Thinking | adaptive, always on | adaptive, always on | adaptive |
| Default effort | high | medium | high |
| Knowledge cutoff | June 2026 | June 2026 | June 2026 |
The window and the output limit are the same across the three. The differences are price, latency and what each model is meant for.
Who needs Fable 5.2 and who can stay on Opus 5.5 or Sonnet 5.5
For most coding and knowledge tasks Anthropic recommends starting with Opus 5.5, which it calls its strongest daily driver, then moving up to Fable or down to Sonnet 5.5 and Haiku 5.5 depending on scope. Our guide Claude Opus 5.5 vs Sonnet 5 vs Fable 5.1: which API model to pick suggests testing the smallest model that meets your acceptance checks, keeping Sonnet for everyday work and trying Fable only after a small evaluation on your own prompts. It was written on October 3, before Fable 5.2 launched, and compares Fable 5 and 5.1, but the order is the same: you reach for Fable after testing the cheaper models on your tasks, not as the default.
| Task | Model | Basis |
|---|---|---|
| Multi-hour sessions, multistep research, long autonomous agents | Fable 5.2 | typical uses on the model page |
| Most coding and knowledge tasks | Opus 5.5 | Anthropic's recommended starting point |
| Everyday code and agents | Sonnet 5.5 | Anthropic's description: best combination of speed and intelligence |
| Classification, extraction, routing | Haiku 5.5 | model for high-volume, low-latency work |
| Advisor or orchestrator for smaller models | Fable 5.2 | Anthropic names both roles |
One caution for privacy-sensitive teams: by the documentation, Fable 5.2 requires 30-day data retention and is not available under zero data retention (ZDR) unless Anthropic expressly authorizes it. If you have strict retention requirements, settle the question before migrating.
For how the bill is structured, see Claude API pricing. If you still call Sonnet 4.5, read what to do when Claude Sonnet 4.5 retires.
How to call Fable 5.2 through the API
In the Claude API you call the model by the ID claude-fable-5-2. On Amazon Bedrock the ID is different, global.anthropic.claude-fable-5-2 or us.anthropic.claude-fable-5-2, and only the bedrock-runtime endpoint serves it. The Python request below is assembled from the fast mode documentation page: the code there uses claude-opus-5-5, and we swapped in claude-fable-5-2, which the same page lists as supported. We did not run this request.
import anthropic
client = anthropic.Anthropic()
response = client.messages.create(
model="claude-fable-5-2",
max_tokens=4096,
speed="fast",
messages=[
{"role": "user", "content": "Refactor this module to use dependency injection"}
],
)
print(response.usage.speed)
Without speed="fast" the same request runs in standard mode. Keep max_tokens generous: thinking is always on in Fable 5.2 and counts toward that limit.
What to watch in requests, by Anthropic's documentation:
- forced
tool_choicetypesanyandtoolreturn a 400, whileautoallows an answer with no tool call: name the tool in the instruction and setstrict: trueon it, which validates the arguments against the schema but does not guarantee a call (in organizations with customer-managed encryption keys strict is not available for Fable 5.2); - do not send
temperature,top_portop_k, and there is a separate article on the 400 these parameters cause; - thinking cannot be turned off, and the
effortparameter controls its depth; - when moving from Fable 5.1, a request that edits the system prompt, tools or earlier messages and then sends old thinking blocks back can get a 400 (the prefix check), and Fable 5.1 cannot read Fable 5.2's thinking blocks;
- Fable 5.2 has its own rate limit, separate from the combined limit shared by Fable 5.1 and Fable 5.
In Claude Code you select Fable 5.2 with /model fable or by launching with claude --model fable. In Claude apps gateway sessions the fable alias resolves to Fable 5 unless ANTHROPIC_DEFAULT_FABLE_MODEL is set, so run /model claude-fable-5-2 there. By Anthropic's documentation, it needs Claude Code 2.1.321 or newer (update with claude update), and Fable models open the 1M window on their own, so the [1m] suffix is not needed. Installation and setup are covered in the Claude Code setup guide. We did not run Claude Code against Fable 5.2.
Fable 5.2 and Claudexia
The list of Claudexia models comes from GET /v1/models, but that response has no prices. Prices are on the models page and in the public pricing API at https://claudexia.tech/api/public/pricing (with channel=premium or channel=cheap). As of November 17, 2026 there are no Fable models in the GET /v1/models response or in the pricing list, so Fable 5.2 cannot be called through the gateway right now.
FAQ
Is Fable 5.2 cheaper or more expensive than Fable 5.1?
At Anthropic it is cheaper: $8 and $40 against $10 and $50 per million tokens. Against Opus 5.5 it costs exactly twice as much: $8 and $40 against $4 and $20.
What is Mythos 5.2 and can I get it?
It is the same model with part of the cyber and biology safeguards removed. Anthropic gives access to organizations that pass verification in its Cyber and Life Sciences programs.
Is Claude Fable 5.2 free?
Fable is not available on the Free plan, and the launch offer did not cover it. Pro subscribers got a week of Fable until November 16, 2026, within 50% of the weekly limit, and it has ended. As of November 17 Fable on Pro runs on usage credits only, while on Max and Ultra it is included up to 50% of the weekly limit. Through the API you pay per token.
Do I need fast mode?
It pays off where you wait for the answer live: IDE edits, interactive debugging. For up to 2.5x faster output you pay twice as much on input and output. For background jobs use standard mode.
What context window does Fable 5.2 have?
By Anthropic's documentation, the 1M-token window is shared: the prompt, the thinking and the response all count toward it. The 128K output maximum per request is not added on top of that million. If input and output together reach the window, generation ends early with stop_reason: "model_context_window_exceeded".