Mythos 5 Context Window: 1M Tokens In, 128k Out

How big is the Mythos 5 context window?

Mythos 5 accepts up to 1M input tokens and can produce up to 128k output tokens per response. Because Fable 5 and Mythos 5 are the same underlying model differentiated only by safeguards, Fable 5 has identical limits.

What 1M tokens of input enables

A 1M token window is large enough to hold, very roughly, hundreds of thousands of words at once. Practical uses include:

• Loading an entire mid-size codebase to reason across files without manual chunking.
• Ingesting long contracts, filings, or research corpora and asking questions across all of them simultaneously.
• Sustaining long, stateful agentic sessions where the full history stays in context.

Why 128k output matters

A 128k output ceiling means the model can return very long artifacts, large refactors, book-length drafts, or exhaustive structured extractions, without truncating mid-task. Most models cap output far lower, so this is a meaningful headroom increase.

A note on cost and retention

Long contexts are powerful but not free: post-intro pricing is $10 per million input tokens and $50 per million output tokens, so a full 1M-token prompt has real cost. Mythos-class also retains data for 30 days with no zero-data-retention option, which matters for sensitive, high-volume context. And note that access to Mythos 5 is Glasswing-only and currently suspended under a June 2026 export-control directive.