Anthropic introduced Claude Fable 5.2 and Claude Mythos 5.2, two variants of the same underlying model aimed at extended technical tasks in coding, science, and engineering. Their main difference is access: Fable includes stronger safeguards for cybersecurity and biology risks and is available to all users, while Mythos is restricted to verified organizations that apply through the company’s Cyber and Life Sciences programs.[1]

The release, dated November 9 by sources tracking the announcement, comes with a price reduction from the previous generation. Anthropic priced both models at $8 per million input tokens and $40 per million output tokens. The company also offers a fast mode, with token generation up to 2.5 times faster, for $16 and $80, respectively. Anthropic says the standard variants generate output up to 30% faster than Fable 5.1 and Mythos 5.1.[1][2]

The most striking demonstration is not a single chat result but a prolonged coordination experiment. Anthropic says Fable 5.2 directed more than 1,000 subagents for eight days to build a working operating system that can boot on a laptop. The company also credits the model with formalizing a proof of Fermat’s Last Theorem in 30 hours. These results point toward a growing ability to divide projects, assign work, and sustain progress over several days, but they remain selected tests disclosed by the developer rather than independent validation of general performance.[2][3]

In prerelease testing, Anthropic says the model helped complete software migrations it had previously considered infeasible, designed three working electronic devices, solved two open problems in black-hole physics, and made progress on robotics evaluations. The wording matters: these are Anthropic-reported outcomes under prerelease conditions. The company’s own post notes that the outside organizations it mentions did not review the announcement and do not necessarily endorse its claims.[1]

Fable and Mythos share the same core, so their separation acts as a risk-based distribution policy rather than two different tiers of intelligence. Anthropic says Fable’s cybersecurity and biology safeguards are as robust as those in Fable 5.1 while producing fewer unnecessary refusals. For Mythos, organizations must demonstrate a beneficial use in biomedical or security research before gaining access. The company also says it added its strongest protections yet against distillation attacks intended to extract model capabilities.[1]

Safety work included evaluations of behaviors such as sandbox escapes. Anthropic says high-stakes risks from misalignment remain low and that the model matched or exceeded recent versions on most internal alignment measures. That conclusion should be read as the provider’s assessment: the announcement alone does not turn internal testing into independent evidence of safety in real-world deployments.[1]

The material development is the combination of three elements: long-duration technical work, large-scale agent coordination, and differentiated access based on risk. If the reported results can be reproduced outside Anthropic, Fable 5.2 could broaden everyday use of agents for software and engineering projects, while Mythos could provide a controlled channel for more sensitive capabilities. The decisive test will be whether external teams achieve comparable gains at sustainable cost and whether the safeguards hold when the model operates on real systems rather than only in evaluations prepared for launch.[1][2][3]

Sources

[1] Read More — Introducing Claude Fable 5.2 and Claude Mythos 5.2 — Anthropic
[2] Read More — Anthropic's New Model Built an OS With 1,000 Agents — Artificially Intimidating
[3] Read More — Anthropic ships Fable 5.2 and Mythos 5.2 — AI Weekly