Anthropic launches new Claude Fable 5.2, Mythos models before IPO

On November 9, Anthropic introduced two novel AI models, Claude Fable 5.2 and Claude Mythos 5.2. The company, based in San Francisco, aims to strengthen its standing against OpenAI before its anticipated initial public offering. Fable 5.2 is tailored for coding, reasoning, and AI agent tasks, while Mythos 5.2 caters to approved cybersecurity and life-sciences research.

However, access to each model varies; Mythos 5.2 remains limited to organizations cleared through Anthropic's verification processes for cybersecurity and life-sciences work. In contrast, Fable 5.2 can be accessed by a wider range of users, with developers able to interact with it via the Claude API using the identifier claude-fable-5-2.

In essence, Anthropic claims a 30 percent faster output compared to the previous Claude 5.1 models, alongside enhancements in coding, reasoning, and vision, positioning Fable 5.2 as the benchmark in coding, finance, and the sciences. Pricing for the models has also been reduced, with input tokens costing $8 per million, and output tokens at $40 per million, both lower than previous releases.

Both models share the same foundational system but Fable 5.2 incorporates additional safety measures for biology and cybersecurity. A system card accompanying the launch outlines safety tests for both models. Positioned above Opus in Anthropic’s model lineup, this launch coincides with the company's IPO roadshow, where investors are assessing enterprise demand and valuation risks.

The 5.2 upgrade was introduced five months after the initial Fable and Mythos releases. Anthropic cautions about potential catastrophic AI risks ahead of its $100 billion IPO.