Claude Mythos 5

Claude Mythos 5 went out on 9 June 2026 alongside Claude Fable 5. The two share the same underlying model, the same specifications and the same price; the difference is what sits in front of them. Fable 5 ships with safety classifiers that can decline a request outright, and Mythos 5 does not include them. For that reason it is not generally available: it is offered in limited release to approved customers in Anthropic's Project Glasswing, reached through an Anthropic, AWS or Google Cloud account team, and it succeeds Claude Mythos Preview. The specifications match its sibling — a one-million-token context window, up to 128,000 output tokens per request, text and image input with text output, ten dollars per million input tokens and fifty per million output. Adaptive thinking is always on and cannot be switched off; depth is controlled with the effort parameter, and the raw chain of thought is never returned, only a summary or an empty block. Task budgets, the memory tool, code execution, programmatic tool calling, context editing, compaction and vision are all supported. Both models carry thirty-day data retention and are designated Covered Models, so the zero-retention option does not apply. The unfiltered configuration is the point and also the risk. Anthropic positions the model for infrastructure providers and vetted security researchers, and published round-ups rank it at the top for finding and exploiting software vulnerabilities. That capability is precisely what interrupted the launch: on 12 June 2026 Anthropic suspended access to both models after the United States government applied export controls, which required restricting access by nationality — something the company could not verify in real time. Access returned on 1 July 2026 once the controls were lifted. Anyone without Glasswing access is directed to Claude Fable 5, which Anthropic describes as offering the same capabilities.

▸News

▸Videos

No videos yet.

▸Reviews