Anthropic Debuts Claude Fable and Mythos 5.2 to Tackle Frontier AI

Anthropic debuted its new Claude Fable 5.2 and Mythos 5.2 models on Monday, positioning the releases as its direct answer to OpenAI's GPT-6 Astra, according to Fortune. The company states that the updated models are designed to tackle daunting technical challenges, showing notable improvements in coding, reasoning, and vision while generating outputs faster than their 5.1 predecessors, The Verge reports. The launch arrives as Anthropic seeks to re-accelerate revenue growth following a third-quarter slowdown noted in its SEC filings.

Pricing plays a central role in the new release. Fortune reports that Fable 5.2 is 20% cheaper than Astra for standard input and output tasks, priced at $8 per million input tokens and $40 per million output tokens compared to OpenAI's $10 and $50 rates. For cached prompts, Anthropic undercuts the competition by 80%, charging 20 cents per million cached tokens against Astra's dollar. TechCrunch notes that Fable is also compatible with zero data retention policies through the Enterprise Frontier Safeguards system.

In pre-release testing, Anthropic claims the models helped engineers complete complex software migrations, designed three working electronic devices from scratch, and made progress on robotics evaluations, according to The Verge and Fortune. Furthermore, the models reportedly solved two open problems in the physics of black holes and formalized the proof of Fermat’s Last Theorem in 30 hours using a team of agent frameworks. TechCrunch adds that startup Physical Superintelligence announced a proof of the Aaronson-Ambainis conjecture using a non-public Anthropic model.

The models also venture into cybersecurity and specialized research. Fortune notes that Claude Mythos 5.2 was found by an unnamed security company to identify vulnerabilities missed by other frontier models, though Anthropic includes safeguards blocking high-risk cyber activities and research biology tasks. TechCrunch highlights that the release follows a METR report touching on recursive self-improvement concerns, while Anthropic's system card details that Mythos 5.2 shows strong covert capabilities on specific evaluations.

Availability varies by model and tier. The Verge and Fortune report that Claude Fable 5.2 is available immediately across platforms including Amazon Web Services, Google Cloud, and Microsoft Azure. Meanwhile, Mythos 5.2 access remains strictly application-only for vetted organizations, cybersecurity professionals, and life sciences researchers, with enterprise customers also able to access it via Claude Security.