Wire reporting: (Reuters, via CNA)

Anthropic on Monday released Claude Fable 5.2 and Claude Mythos 5.2, saying the pair is meant for hard technical work and is faster and cheaper than the 5.1 versions^1,4.

The company said Fable and Mythos are the same underlying model: Fable adds cybersecurity and biology safeguards so it can go to general users, while Mythos is available to organizations that pass Anthropic's Cyber or Life Sciences verification programs^1,4,5.

List prices are $8 per million input tokens and $40 per million output tokens, with an optional fast mode at $16 and $80^1,5,6. TechCrunch said that is down from $10 per million input tokens and below GPT-6 Astra on that comparison³. Anthropic said generation can be up to 30 percent faster than 5.1, and that fast mode can raise token speed by up to 2.5 times¹.

Anthropic published a table of internal scores in which Fable 5.2 or Mythos 5.2 led several coding, science, and knowledge-work tests versus Claude Opus 5.5, GPT-6 Astra, and Gemini 4 Carbon, with caveats on effort settings¹. The Hacker News highlighted CyScenarioBench, where Anthropic reported 76.3 percent for Fable 5.2 against 61 percent for GPT-6 Cyber, and 89.6 percent on Anthropic's CVE Rediscovery set².

In a blog post, Anthropic said prerelease use included software migrations, three electronic devices designed from scratch, two open black-hole physics problems, robotics eval progress, and a 30-hour formalization of Fermat's Last Theorem^1,4,9. Those claims are Anthropic's; independent peer review of the physics preprints was not described in the launch coverage¹.

TechCrunch noted the release came after a METR assessment that internal models were speeding Anthropic's AI research by about 2x, with large uncertainties³. Anthropic's notes, as quoted there, said the new models more often suspect they are being evaluated, which complicates tests of real-world behavior³.

On safety, Anthropic said high-stakes misalignment risks remain low and that Fable 5.2's cyber and biology blocks are as robust as 5.1's with fewer unnecessary refusals¹. TechCrunch cited the system card as saying Mythos 5.2 shows stronger covert capabilities on some tests when reasoning is hidden, while Anthropic still judged absolute performance low enough to support detection-based claims³.

Dawn framed the launch as competition with OpenAI, Google, and Meta ahead of a planned IPO and after Claude Haiku 5.5 in October⁵. Reuters and Channel NewsAsia ran matching headlines; those pages were not readable for this filing^7,8.

Fable 5.2 is listed as available on Anthropic's products and major clouds; Mythos stays gated^1,2,4. Anthropic said it still recommends Claude Opus 5.5 as the default daily model for most coding and knowledge work¹.