On November 9, Anthropic introduced versions 5.2 of Claude Fable and Claude Mythos, its models designed for the most demanding technical work. The publisher claims improvements in programming, reasoning, and vision, as well as response generation up to 30% faster than versions 5.1.
One model, two access levels
Fable and Mythos are based on the same underlying model. However, Fable incorporates enhanced safeguards in the areas of cybersecurity and biology, which allows Anthropic to open it to all users. Mythos, on the other hand, is reserved for organizations conducting work in cybersecurity or biomedical research.
Both models are priced at $8 per million input tokens and $40 per million output tokens, a token being a unit of text, often a fragment of a word. An optional fast mode, which speeds up generation by up to 2.5 times, increases these rates to $16 and $80. Anthropic nevertheless recommends starting with Claude Opus 5.5 for most development and knowledge tasks.
Claude Fable 5.2 has been available since November 9 on all platforms, including Amazon Web Services, Google Cloud, and Microsoft Azure, as well as on the Claude Platform, under the identifier claude-fable-5-2. Eligible paid plan subscribers receive additional usage, with one to three resets depending on their tenure. Fable can account for up to 50% of the weekly usage limits for Pro plans and standard seats on the Team plan.
Cybersecurity professionals can request access to Mythos 5.2 via the Cyber Verification Program, and life sciences researchers via the Life Sciences Verification Program. Claude Enterprise customers can use Claude Security, which leverages all of the publisher’s models, including Mythos 5.2. The fast mode, offered on Claude Code, the Claude Platform, Google Cloud, and Microsoft Foundry, remains for Mythos restricted to organizations that are members of these programs.
Increased performance on code
According to the publisher, on FrontierCode 2.0, Fable 5.2 achieves 49.1%, compared to 32.4% for Fable 5.1, 45.5% for Opus 5.5, and 42.8% for GPT-6 Astra. With a high effort level, it matches Opus 5.5 for about half the cost per task. The model does not dominate all tests: on AutomationBench, focused on business workflows, it scores 51.4%, behind Gemini 4 Carbon’s 53.9%.
Anthropic states that over 90% of changes integrated into its main source code are now written by Claude, under human supervision. In an internal experiment, Fable 5.2 coordinated more than 1,000 sub-agents, AI instances working under the direction of a main agent, for eight days. The goal was to produce a prototype operating system capable of booting on a laptop.
Cybersecurity, science, and safety
Anthropic presents Mythos 5.2 as its highest-performing model for cybersecurity engineering. In an internal evaluation covering 1,829 CVEs, identifiers for known vulnerabilities, it identified 10.2% more vulnerabilities than Opus 5.5 and 14.4% more than Mythos 5.1. At Mozilla, Distinguished Engineer Brian Grinstead estimates that Mythos 5.2 has “roughly doubled” the vulnerability discovery metrics of Mythos 5.1.
On the scientific front, according to the publisher, Fable 5.2 solved two open problems in theoretical physics related to black holes. A team of 1,000 Fable 5.2 agents also formalized the proof of Fermat’s Last Theorem in 30 hours, compared to eleven days for a first attempt conducted with an unpublished research version.
Regarding safety, Anthropic judges the risk of dangerous behavior in high-stakes situations to be low. The model was evaluated before its release by the UK AI Security Institute, Frontier Design, and METR. However, most published results come from evaluations conducted by Anthropic.
