Anthropic has launched Claude Fable 5.2 and Claude Mythos 5.2, reclaiming the AI performance lead ahead of the company’s planned initial public offering. Built on the same underlying foundation, Fable 5.2 is publicly accessible with embedded safety guardrails, while Mythos 5.2 removes select restrictions for verified cybersecurity and life sciences organizations. The models carry a base pricing of $8 per million input tokens and $40 per million output tokens, representing a price reduction from their predecessor models.

According to Artificial Analysis benchmarks, Fable 5.2 placed first on the Index v5 compute tier with 65 points, outpacing unreleased models like Google’s Gemini 4 Carbon and Anthropic’s own Opus 5.5. The model demonstrated strong performance in expert academic evaluation (80% on HLE-Diamond) and scientific terminal workflows (73% on Terminal-Bench-Science), though it showed relative weakness in multi-step agentic knowledge work and terminal coding.

Anthropic continues to recommend Claude Opus 5.5 for standard daily operations, positioning Fable and Mythos for complex technical tasks. While output token consumption efficiency has improved in certain tiers, peak compute usage brings average per-task costs to $26.50. The update also includes an optional high-speed mode offering up to 2.5 times faster output generation at double the standard cost across select enterprise platforms.

Why it matters

  • Fable 5.2 sets a new benchmark for technical tasks, though high compute costs ($26.50/task at peak tier) require careful token management.

  • Mythos 5.2 provides restricted, high-capability access for specialized biotech and security work via Anthropic’s verification program.

  • Anthropic is establishing market dominance across specialized technical benchmarks right before its planned initial public offering.

Source: the-decoder.com