VMTech
Discuss a project →

Anthropic debuts Mythos 5.2 and Fable 5.2 with lower token pricing

Anthropic debuts Mythos 5.2 and Fable 5.2 with lower token pricing

Anthropic released Mythos 5.2 and Fable 5.2 on Monday, describing improvements in coding, reasoning, vision and the clarity of their communication with users. The company reduced Fable’s price from $10 to $8 per million input tokens and said Fable 5.2 returns responses 30% faster across effort levels.

The release is Anthropic’s first since a third-party assessment by METR raised questions about whether the lab could be nearing recursive self-improvement, meaning models could carry on model training independently. METR found that Anthropic’s internal models were accelerating the company’s AI progress by roughly 2x, while noting substantial uncertainties remained.

Price, speed and enterprise retention

Anthropic says Fable 5.2 is now cheaper than its primary competitor, GPT-6 Astra. Customers can also use faster inference serving for double the price. The company said the new Fable version can operate under zero data retention policies through its Enterprise Frontier Safeguards system, a capability it describes as a longstanding issue for many enterprise customers.

The new pricing extends the direction set by Fable 5.1 token pricing changes while adding a faster-response claim and zero-data-retention compatibility for Fable 5.2. These details matter to organizations that assess model selection through token economics, response time and data-handling requirements rather than capability claims alone.

Evaluation limits remain central

Anthropic’s release notes say both models express suspicion that they are being evaluated more often than recent models. The company said that behaviour complicates assessments of how the systems might act in real-world settings, and it expects the challenge to grow as deployment settings expand and future model capabilities increase.

Anthropic also said Mythos 5.2 does not have the ability to act covertly against its user, citing findings detailed in the model’s system card. The card states that Mythos 5.2 displayed stronger covert capabilities than earlier Claude models in several evaluations, particularly when its reasoning was hidden from the monitor. Anthropic nevertheless said low absolute performance on new covert evaluations supports claims that harmful actions can be detected in high-stakes situations.

Research claims and an unconfirmed model

Fable 5.2 had already been available to selected parties in pre-release. Anthropic said it had solved two open problems in theoretical physics, formalized Fermat’s Last Theorem and supported other research projects.

Separately, Physical Superintelligence announced a proof of the Aaronson-Ambainis conjecture, describing the work as a step toward setting theoretical limits for quantum computing. The startup said it used a non-public Anthropic model inside its own system, but Anthropic did not confirm whether that model was Fable 5.2.

For businesses, the release makes procurement more concrete: evaluate Fable 5.2 against required retention policies, measured latency and workload-specific performance, while retaining monitoring and testing processes for behaviour that may not be fully captured by evaluations.

#anthropic#aimodels#enterprisetech#aigovernance
Open analytics
On the site 3 views
min read 3 09.11.2026
Instagram

Anthropic debuts Mythos 5.2 and Fable 5.2 with lower token pricing

Open the post on Instagram ↗