Anthropic's New Models Have Beat all other Models

Fable 5.2 is pitched as Anthropic's most capable model for "ambitious, open-ended technical work": multi-repo coding that tests its own output, reasoning that runs for days and coordinates its own subagents, and vision sharp enough to read a cluttered chart at nearly double its predecessor's accuracy. Anthropic says it beats its own Opus 5.5 on several benchmarks, alongside claimed wins over GPT-6 Astra and Gemini 4 Carbon, and dropped API pricing from $10/$50 to $8/$40 per million input/output tokens, with a faster mode for people willing to pay more to wait less. Mythos 5.2 is the same underlying model minus Fable's cybersecurity and biology guardrails, available only to vetted researchers through Anthropic's Life Sciences and Cyber Verification programs, which is a tidy way of selling the exact same car with the governor chip removed, but only to people who can prove they will not drive it into a hospital.

The naming is the actual brand story. Where OpenAI ships "GPT-6" and Google ships "Gemini," Anthropic keeps building a mythology-coded family (Opus, Sonnet, Haiku, now Fable and Mythos) that frames its models as creative collaborators rather than spreadsheet tools, even while the underlying pitch is agentic coding and quantitative finance. That is a deliberate wedge against competitors selling raw horsepower: Anthropic is selling a worldview where the machine is a storyteller first and a coder second, which matters a lot more to enterprise buyers deciding whose AI "voice" they want baked into their own products for the next five years.

Source: Anthropic

Kyle Duford, Founder & Editor, BRNDWIRE. More from Kyle →