AI Sunday
EN

Claude Fable 5.2: 20% cheaper, Mythos 5.2 by application

2–4 minutes

Claude Fable 5.2 has been available since Monday, November 9, and it costs less than the previous version. Its twin Mythos 5.2 remains reserved for verified organisations.

Fable 5.2 for everyone, Mythos 5.2 by application

Anthropic says it plainly: Fable and Mythos are the same model. Fable ships with cybersecurity and biology safeguards, which makes it possible to open it to everyone. Mythos partly lifts them. To get it, an organisation must apply to the cybersecurity or life-sciences verification programme.

Fable 5.2 is available to Pro, Max, Ultra, Team and Enterprise subscribers, and through the API under the name claude-fable-5-2. Amazon Web Services, Google Cloud and Microsoft also offer it. On the Mythos side, the Claude Security tool now runs on it.

Fable 5.2Mythos 5.2
Who has accessPaid subscribers and APIVerified organisations
Cyber and biology safeguardsActivePartly lifted
Price per million tokens (input / output)$8 / $40From $8 / $40
Both models rest on the same technical foundation. Source: Claude Fable and Claude Mythos pages, Anthropic.

A 20% price cut

Fable 5.2 charges $8 per million input tokens and $40 for output. Fable 5.1 cost $10 and $50. A token is a small piece of a word, the unit on which the API counts its bill. The drop therefore shows up first on the bills of developers and companies that go through the API.

Graphique en barres : prix de Claude Fable par million de tokens, en dollars. Fable 5.1 : 10 en entrée, 50 en sortie. Fable 5.2 : 8 en entrée, 40 en sortie. Fable 5.2 en mode rapide : 16 en entrée, 80 en sortie. Source Anthropic, 9 novembre 2026.
The optional fast mode speeds up generation by up to 2.5 times for double the price. Source: Claude Fable page, Anthropic.

The model also responds up to 30% faster than Fable 5.1. In a first test on FrontierCode 2.0, Fable 5.2 at “High” effort matches Opus 5.5 at “Max” effort, for about half the cost per task. These tests are Anthropic’s own.

What Anthropic claims to have measured

On FrontierCode 2.0, a programming test, Fable 5.2 scores 49.1%, versus 45.5% for Opus 5.5 and 32.4% for Fable 5.1, according to Quartz. On RoboDojo, a robot-control test, it goes from 27.6% to 64.9%, according to The Next Web. Anthropic adds that more than 90% of the changes to its main code are now written by Claude, with human guidance.

30 hours

to formalise the proof of Fermat’s last theorem with 1,000 Fable 5.2 agents, versus 11 days for Anthropic’s first attempt

The limits of the launch

None of these results has yet been reproduced by a third party. Anthropic also acknowledges that both models say they suspect they are being tested more often than previous ones, which makes their behaviour in real situations harder to predict, according to TechCrunch, citing the safety card.

Another concrete point for a company: Fable models impose 30 days of data retention by default for safety monitoring. Customers who require zero retention will have to wait for the Enterprise Frontier Safeguards option.

What this changes at work

For most employees, nothing tomorrow morning. Anthropic itself advises starting with Opus 5.5 and only moving up to Fable or Mythos for the most open-ended technical work. Development, data and finance teams are the first affected by the price cut. Security teams can apply for Mythos.

This release comes as Anthropic prepares its stock-market debut, for which we detailed what the prospectus reveals. At the end of October, we also looked back at the training pause of the next model.

Sources

À découvrir aussi sur AI Sunday

Portrait de Daniel, fondateur d'AI Sunday

Écrit par

Daniel

Professionnel des achats, j'utilise l'IA au quotidien dans mon travail et je raconte ici ce qui marche vraiment, et ce qui coince. En savoir plus