Anthropic has rolled out Claude Fable 5.2 and Mythos 5.2, the latest versions of its AI models aimed directly at OpenAI's GPT-6 Astra. The San Francisco-based company said the new models are not only cheaper for routine tasks but also cut the cost of processing large cached prompts by as much as 80 percent compared with OpenAI's flagship.
For standard input and output, Fable 5.2 and Mythos 5.2 are priced at $8 per million input tokens and $40 per million output tokens. That is 20 percent below GPT-6 Astra, which costs $10 per million input tokens and $50 per million output tokens. For cached prompts, Anthropic's models cost 20 cents per million tokens, compared with $1 per million for Astra.
A coding and research push
Anthropic said its new models outperform rivals on coding benchmarks, a category where the company has built a strong following through its Claude Code assistant. The company noted that more than 90 percent of its own software is now written by AI, underscoring how heavily it relies on its own tools.
The announcement also highlighted a research milestone. A team of 1,000 agents running on Fable 5.2 formalized a mathematical proof in roughly 30 hours, down from 11 days for an earlier unreleased Claude research version. Formalization converts proofs into a machine-checkable format, a task that normally takes skilled mathematicians far longer.

