AI Model Timeline

Tracking the accelerating release frequency of frontier AI models.

Claude Mythos 5.2

Anthropic·

LLMsflagshipcloud

Overview

Same underlying model as Claude Fable 5.2, with the cyber and biology safeguards that govern Fable 5.2 lifted. Successor to Claude Mythos 5.1. Access is limited to organizations vetted through Anthropic's expanded Life Sciences and Cyber Verification Programs; Claude Enterprise customers can also reach it through Claude Security. Shares Fable 5.2's pricing ($8/M input, $40/M output; fast mode $16/$80), 1M context and 128K max output. Anthropic calls it its most capable model for cybersecurity engineering and reports only untracked evaluations for it: CVE Rediscovery 89.6 (Mythos 5.1: 78.3), CyScenarioBench 76.3, Terminal-Bench-Science 0.1 75.3. Independent evaluators measure the public Fable 5.2 endpoint, not this variant, so no independent values are carried over.

Capabilities and innovations

Autonomous long-horizon tasksAgentic codingCyber defenseMultistep researchComputer useVision inputAPI: claude-mythos-5-2Life Sciences and Cyber Verification Program accessAvailable inside Claude Security for Enterprise customers

Benchmarks

No benchmark values are recorded for this model.

  • GPQA Diamond: independent evaluation pending — Gated model; independent boards (Vals AI, Artificial Analysis, ARC Prize) evaluate the public Fable 5.2 endpoint, not this variant. Fable 5.2 values are not copied over: the 5.1 launch showed the safeguards move scores (Terminal-Bench 4.0 60.9 vs 55.8).
  • Humanity’s Last Exam (no tools): independent evaluation pending — Anthropic's launch table reports HLE (74.2 with tools) for Fable 5.2 only, not for Mythos 5.2.
  • SWE-bench Verified: not reported by the vendor
  • SWE-bench Pro: independent evaluation pending — Not on the announcement page; system card unreachable from the intake sandbox.
  • MMLU: not reported by the vendor
  • MMLU-Pro: independent evaluation pending — Gated model; no independent measurement of this variant.
  • Terminal-Bench 4.0: independent evaluation pending — No Mythos 5.2 Terminal-Bench 4.0 figure on the announcement page (Mythos 5.1 had one); system card unreachable from the intake sandbox.
  • MMMU-Pro: independent evaluation pending — Gated model; no independent measurement of this variant.

API pricing

$8 per million input tokens, $40 per million output tokens (USD, provider list price, checked 2026-11-09)

Links

More from Anthropic

Data curated by AI Model Timeline. See the methodology for admission criteria, benchmark eras and source priority.