Claude Fable 5.2 vs. Mythos 5.2: differences and pricing
Anthropic introduced Claude Fable 5.2 and Mythos 5.2 with the same base model and different access conditions. The update lowers input and output rates, adds a fast mode, and keeps restrictions for specialized work.
Anthropic introduced Claude Fable 5.2 and Claude Mythos 5.2 on November 9, 2026, with announced improvements in coding and reasoning. The two share the same base model, but their safeguards and access conditions are different, according to Anthropic's announcement.
Fable is the general-access configuration, subject to the corresponding plans and platforms. Mythos is aimed at verified organizations working in areas such as cybersecurity and biomedical research. That distinction is central to understanding the launch: contracting access to Claude does not automatically enable all of Mythos's capabilities.
The difference between Fable and Mythos
The Fable 5.2 documentation describes capabilities and specifications shared with Mythos. It does not present Mythos as a version that, by its name, is superior at any task.
Fable includes additional filters for sensitive cybersecurity and biology activities. Mythos allows certain work in those areas for organizations that have passed Anthropic's verification programs.
An example helps separate the uses: modifying an inventory system and conducting vulnerability research require different permissions. Both may involve programming, but the second falls into a category with specific controls.
Verified access should not be interpreted as an absence of rules either. It corresponds to a modality intended for organizations and approved uses.
What this update is designed for
Anthropic positions Fable 5.2 for demanding work that can extend over hours and needs to chain multiple steps. The documentation mentions deep research and analysis that end in documents or other complete deliverables.
In everyday terms, it targets tasks where a brief answer is not enough: gathering information, working on it, and maintaining continuity until producing a result.
The what's new guide maintains a relevant condition: for most workloads, it suggests starting with the latest Opus model and turning to Fable when the results do not meet the requirements.
The launch expands the options for difficult work. It does not establish that all queries should move to the most expensive model in the family.
How much they cost in the API
The API is the connection that lets you embed Claude inside an application. Its rates are based on model usage, not on the monthly price of a chat subscription.
Fable 5.2 and Mythos 5.2 share these published rates in U.S. dollars:
| Mode | Per million input tokens | Per million output tokens |
|---|---|---|
| Standard | US$8 | US$40 |
| Fast mode | US$16 | US$80 |
| Batch processing | US$4 | US$20 |
Tokens are pieces of information that the system processes or generates. Input includes the material sent; output corresponds to what is produced.
The previous generation had standard rates of US$10 and US$50, respectively. That's a 20% reduction on those two rates, calculated from the prices published in the release notes documentation.
That doesn't guarantee any project will cost 20% less. The total depends on how much material the system processes and generates, as well as the tools and modes used.
Fast mode charges for higher speed
Anthropic announces up to 2.5 times faster output generation with fast mode, in exchange for double the standard rates. The cost optimization documentation explicitly distinguishes that trade-off.
The figure refers to output tokens per second. It does not prove that a research task or a software change finishes 2.5 times sooner.
A complete job can include searches, tool execution, and reviewing results. Total wait time includes those stages, in addition to text generation.
Fast mode also needs to be separated from effort level. The first changes output speed; the second controls how much processing the model devotes to handling the request.
What capabilities they share
The Fable 5.2 spec sheet sets a context window of one million tokens and a maximum output of 128 thousand tokens. It accepts text and images as input and produces text.
The context window describes how much material it can consider within a request or conversation. It is not the same as permanent memory, nor does it guarantee that every detail gets the same attention.
Adaptive reasoning stays on: the system decides how to distribute its processing, while the effort setting lets you adjust its depth. It's a feature shared by both configurations.
The promotion has an end date and limits
Anthropic announced a launch promotion for certain paid plans. Pro and standard Team users get one week of Fable models included, up to 50% of their weekly limits.
The help center sets the end of that window at November 16, 2026, at 11:59 p.m. Pacific Time. After that, Fable goes back to consuming usage credits when they're enabled.
Limit resets have eligibility and expiration requirements. They restore available usage; they don't raise your contracted limits. The promotion also doesn't automatically change the selected model.
So included temporary access and unlimited usage are different conditions. The offer should be read according to each account's plan and billing tenure.
Conditions that affect integrations
The migration guide states that both models require 30-day data retention. Zero-retention agreements need express authorization from Anthropic to be used.
There are also compatibility changes. Some integrations that modify previous messages or use older computer control tools may stop working as they did before.
The Fable 5.2 prompting guide acknowledges behavioral limits: with lower effort, the model may stop before finishing a task. Delegating to subagents can also account for a considerable share of usage.
These observations are part of the vendor's documentation. They help clarify what a capability improvement means without turning it into a promise of infallible execution.
Fable expands access; Mythos keeps verification
The concrete change combines lower standard rates with new speed options for demanding workloads. The separation between Fable and Mythos is still determined by safeguards and use authorization.
For those already using Claude, the scope depends on where you consume it: a chat subscription, an API integration, and a verified access program each have different terms. Version 5.2 does not eliminate those differences.
Terms in this note
- Base model: the underlying artificial intelligence system to which different configurations are applied.
- API: a connection that allows Claude to be used from another application.
- Token: a unit of information that the model processes or generates; it does not necessarily correspond to a full word.
- Context window: the maximum amount of information the model can consider in a request.
- Adaptive reasoning: an operating mode in which the model adjusts its processing to tackle a task.
- Batch processing: a mode for submitting groups of requests without requiring an immediate interactive response.
- Data retention: the period during which the provider keeps information in accordance with the terms of service.
- Auxiliary agent: an additional artificial intelligence instance that is assigned part of the work.
References
Query performed on November 9, 2026. The official pages cited offered indexed content, but their full opening was not available in this query. No independent testing of the models or full review of the safety report was performed. Performance figures correspond to Anthropic's claims.
- Anthropic: Claude Fable 5.2 and Mythos 5.2
- Claude Platform: Fable 5.2 overview and specifications
- Claude Platform: what's new in Fable 5.2 and Mythos 5.2
- Claude Platform: optimizing for cost and capability
- Anthropic: Fable 5.2 launch offers
- Claude Platform: migration guide to Fable 5.2 and Mythos 5.2