On 9 November 2026, Anthropic released Claude Fable 5.2 and Claude Mythos 5.2, priced at 8 dollars per million input tokens and 40 dollars per million output tokens, 20% less than the Fable 5.1 rates reported by the news outlet Next. The general-availability model remains restricted on offensive cybersecurity and biology, but blocks far fewer legitimate requests. For businesses, the question is less raw performance than proper use: Anthropic itself does not recommend it as an everyday model.
Two names for the same model
According to Anthropic’s announcement, Fable 5.2 and Mythos 5.2 are based on the same model. Fable includes enhanced safeguards in cybersecurity and biology, which allows it to be made available to everyone. Mythos, which is less restricted, is reserved for organisations approved through the cyber verification programme or the one dedicated to life sciences; Claude Enterprise customers can nevertheless use it through the Claude Security tool without joining the cyber programme.
Fable 5.2 is available from launch on Anthropic’s API and with the three major cloud providers (AWS, Google Cloud and Microsoft Azure). On the subscription side, Next reports that it is available from the Pro plan upwards; as part of a launch promotion, the company specifies that it can account for up to half of weekly usage limits on Pro plans and standard Team seats.
A price cut, and a fast mode that costs twice as much
According to Next, Fable 5.1 cost 10 dollars per million input tokens and 50 dollars per million output tokens. The move to 8 and 40 dollars therefore represents a cut of around one fifth. A fast mode, which speeds up generation by up to 2.5 times, is billed at double the price: 16 dollars for input and 80 dollars for output.
Anthropic mainly highlights an efficiency gain. On the FrontierCode 2.0 programming test, Fable 5.2 at “high” effort scores 45.3%, on a par with Opus 5.5’s best score (45.5%), for around half the cost per task, according to the company. The official recommendation nevertheless remains to use Opus 5.5 for most coding and office tasks, to reserve Fable for the most difficult work, or to step down to Sonnet 5.5 and Haiku 5.5 for simple tasks.
Performance to be put into perspective
The tables published by Anthropic place Fable 5.2 ahead of its competitors GPT-6 Astra and Gemini 4 Carbon in most tests, but not all: on AutomationBench, which measures business process automation, Gemini 4 Carbon keeps the edge (53.9% versus 51.4%). These figures are the company’s own. Next reports that an independent evaluation by Artificial Analysis confirms the model’s lead on long reasoning, with more mixed results on targeted tasks.
The spectacular demonstrations showcased, such as a prototype operating system built in eight days by more than a thousand sub-agents, cannot be verified: according to Next, neither the code of this system nor the formalisation of the proof of Fermat’s Last Theorem, completed according to Anthropic in 30 hours by a thousand agents, has been published.
What changes for businesses
- Fewer unjustified refusals: Anthropic reports, based on its initial tests, around 55% fewer blocks than with Fable 5.1, including 60% fewer in cybersecurity. Blocked requests are redirected to older models, Opus 4.8 for cyber and Opus 5 for biology.
- Data retention: safety monitoring requires 30-day retention by default. A system that works without retention, by analysing exchanges within the customer’s infrastructure, is still only in private testing with a few large accounts.
- Security: in an adversarial test run by Anthropic, Mythos 5.2 used credentials or files placed outside its sandboxed environment in 3.1% of scenarios, a rate that reached 21% with Mythos 5.1 and 39% with Mythos 5. External evaluators, including the UK AI Security Institute and METR, tested the model.
One unknown remains for European customers: the announcement mentions neither the size of the context window nor any option for data residency in the EU, two points to check before any deployment subject to the GDPR.
Frequently asked questions
How much does Claude Fable 5.2 cost?
8 dollars per million input tokens and 40 dollars per million output tokens, in standard mode. Fast mode costs 16 and 80 dollars. Fable 5.1 was priced at 10 and 50 dollars, according to Next.
What is the difference between Fable 5.2 and Mythos 5.2?
It is the same model. Fable is open to everyone with filters on offensive cybersecurity and biology; Mythos, which is less restricted, is reserved for organisations approved through Anthropic’s verification programmes.
Should Opus 5.5 be replaced with Fable 5.2?
Not necessarily. Anthropic recommends Opus 5.5 for most everyday uses and reserves Fable for the most complex tasks. The choice depends on the cost per task actually measured on your own use cases.
Sources
- Introducing Claude Fable 5.2 and Claude Mythos 5.2 — Anthropic, 2026-11-09
- Anthropic relance la course avec Claude Fable 5.2 et Mythos 5.2 — Next, 2026-11-10