Anthropic debuts Claude Fable, Mythos 5.2 models, priced 20% below OpenAI’s GPT-6 Astra

Anthropic unveiled Claude Fable 5.2 and Mythos 5.2 models on Monday, positioning them as alternatives to OpenAI's GPT-6 Astra. These new models promise higher performance at a lower cost than Astra, catering to customers who often switch models based on price. Fable 5.2 and Mythos 5.2 are 20% less expensive than Astra for regular input and output tasks, costing $8 per input token and $40 per output token, compared to $10 and $50 for GPT-6 Astra, respectively.

For tasks involving large amounts of data, such as documents, the cost advantage widens, with Mythos 5.2 being 80% cheaper than Astra at $0.20 per million cached tokens.

During pre-release testing, Anthropic reported that the new models helped engineers with software migrations, designed functional electronic devices, solved physics problems related to black holes, made progress on robotics evaluations, and formalized a proof of Fermat's Last Theorem in 30 hours. The company claims over 90% of its code is generated by AI, a significant advantage for programmers.

While OpenAI's Codex has gained ground in recent months due to lower pricing, Anthropic's models are best suited for complex, open-ended technical work, competing with Astra at the cutting edge. However, for routine tasks, Anthropic recommends its Opus 5.5 model, which is pricier than OpenAI's GPT-6.1 Sol but may offer better value if it consistently produces high-quality results.

Anthropic is also focusing on cybersecurity, claiming one in six bugs found by Mythos 5.2 were missed by other models, and has demonstrated the models' ability to design functional electronic devices from a single prompt. They claim superiority in mathematics, formalizing a 90-year-old proof in 30 hours. For cybersecurity professionals and life sciences researchers, access to Mythos 5.2 is restricted via Anthropic's Claude Security product.