One million tokens of text written by Claude Fable 5.2 cost $40, against $50 for the previous version. That is the figure that stands out in the Anthropic announcement of Monday, November 9, 2026, which introduces two models at once: Fable 5.2 and Mythos 5.2. A token is a fragment of a word, the unit the API (programmatic access to the model) bills on.
According to Anthropic, the two names refer to the same base model. Claude Fable 5.2 gets stronger safeguards in cybersecurity and biology, which makes it possible to open it to everyone. Mythos 5.2 has fewer, and is only available to organizations that apply.
Claude Fable 5.2: $8 for input, $40 for output
Anthropic announces $8 per million input tokens and $40 per million output tokens, a lower price than the 5.1 series, with responses generated up to 30% faster. Fable 5.1 cost $10 and $50, as TechCrunch recalls: the cut is 20%, which CNBC confirms. Our article on GPT-6 Astra and Fable 5.1 details that starting price.
| Model | Input (per million tokens) | Output (per million tokens) |
|---|---|---|
| Fable 5.1 | $10 | $50 |
| Fable 5.2 | $8 | $40 |
| Fable 5.2, fast mode | $16 | $80 |
| GPT-6 Astra (as reported by Fortune) | $10 | $50 |
The optional fast mode speeds up generation by up to 2.5 times for double the price. For cache reads (texts already sent that the server keeps in memory), Fortune reports $0.20 per million at Anthropic against $1.00 at GPT-6 Astra, 80% less. That figure comes from Fortune: Anthropic’s announcement gives no cache price. It gives no price in euros either.
Claude Fable 5.2 is offered on Amazon Web Services, Google Cloud and Microsoft Azure, under the identifier claude-fable-5-2 for developers. On the subscription side, Fable 5.2 can use up to 50% of the weekly usage limits of the Pro plan and of standard seats on the Team plan.
Mythos 5.2: access requires an application
Claude Fable 5.2 ships with classifiers (automatic filters) that block high-risk cyber activity, such as penetration testing or exploit generation, and most biology research tasks. A blocked request then gets an answer from Opus 4.8 in cybersecurity or Opus 5 in biology. According to Anthropic’s first tests, these blocks are about 55% less frequent than with Fable 5.1.
For Mythos 5.2, you have to apply to one of two verification programs: the Cyber Verification Program, organized since October 6 into three access tiers with different verification requirements, and the Life Sciences Verification Program, in beta for teams and institutions, with a review of research credentials, security standards and ethical oversight. Claude Enterprise customers can also use Claude Security, which analyzes source code with Mythos 5.2. Mythos fast mode is reserved for organizations in these programs.
The pages for these programs talk about reduced or refined safeguards. Only one option of the life sciences program, reserved for high-risk uses, removes all the safeguards that block biology requests. No general public access to Mythos has been announced.
The published scores, all signed by Anthropic
Anthropic writes that Claude Fable 5.2 and Mythos 5.2 are “a step above Opus 5.5” on almost everything that was evaluated. The announcement publishes a table, partly reproduced here. The scores for Fable 5.2, Fable 5.1 and Opus 5.5 are measured with safeguards on, at the highest effort level available, and the figures are Anthropic’s.
| Test | Fable 5.2 | Fable 5.1 | Opus 5.5 | GPT-6 Astra | Gemini 4 Carbon |
|---|---|---|---|---|---|
| FrontierCode 2.0 (agentic coding) | 49.1% | 32.4% | 45.5% | 42.8% | 41.1% |
| FrontierSWE v2 (agentic coding) | 69.9% | 56.3% | 62.3% | 65.5% | 61.2% |
| AutomationBench (business tasks) | 51.4% | 31.4% | 42.5% | 41.4% | 53.9% |
| Terminal-Bench-Science 0.1 | 74.6% | 52.6% | 58.7% | 68.1% | 64.8% |
| OSWorld 2.1 (computer use) | 89.3% | 83.5% | 87.2% | 79.5% | 81.5% |
| Humanity’s Last Exam (with tools) | 74.2% | 65.6% | 67.7% | 57.2% | not published |
| Chartography (chart reading) | 73.4% | 44.8% | 64.4% | 71.0% | 72.5% |
| AA-Briefcase v1.1 (Elo score) | 1,785 | 1,669 | 1,800 | 1,569 | 1,611 |
Two rows contradict the story of a clean sweep. Gemini 4 Carbon beats Fable 5.2 on AutomationBench (53.9% against 51.4%), and Opus 5.5 stays ahead on the AA-Briefcase Elo score (1,800 against 1,785). On Chartography, the lead over Gemini is under one point. The best FrontierCode result, 49.1%, comes at the “Xhigh” effort level; at “High” effort, Fable 5.2 scores 45.3% against 45.5% for Opus 5.5, for about half the cost per task, according to Anthropic.
Mythos 5.2 has its own rows. On CVE Rediscovery (finding already published security flaws in a project’s code), it reaches 89.6%, against 78.3% for Mythos 5.1. None of these scores comes from an independent test: the announcement cites no external measurement of capabilities.
System card: concealment, suspicion of being tested and gray areas
The system card (the technical document that comes with the model) contains a sentence that TechCrunch quotes: “Claude Mythos 5.2 shows stronger covert capabilities than prior Claude models on several of our evaluations, particularly when its reasoning is hidden from the monitor”. According to the same outlet, the card nonetheless judges that absolute performance levels remain low on these new tests.
The announcement also flags a limit of the tests themselves: Mythos 5.2 and Fable 5.2 say they suspect they are being evaluated more often than any recent Anthropic model, which complicates estimating their real behavior. In a sandbox escape test (a walled-off environment), Mythos 5.2 used credentials placed outside the sandbox in 3.1% of scenarios, against 39% for Mythos 5 and 21% for Mythos 5.1.
TechCrunch links this launch to a report from the evaluation group METR, according to which Anthropic’s internal models would speed up its progress roughly twofold, which feeds the debate on recursive self-improvement (an AI that would take part in training the next one). Anthropic says it limited the capabilities most useful to AI research during training: on accelerator kernel optimization and model architecture improvement, Fable 5.2 stays at roughly the level of Opus 5.5.
According to CNBC, it is the first release of frontier models since Dario Amodei’s call to slow the pace of AI, made in September. Still missing from the announcement: the list of countries where Fable 5.2 is offered, a price in euros and any independent measurement of its capabilities. The S-1 filed on October 26 places this launch just before Anthropic’s IPO. Fortune sees it as one of the company’s last big releases before that step.
0 Commentaires
Aucun commentaire pour le moment. Soyez le premier à commenter !