Mythos

The 📝Claude Mythos context window is 1 million input tokens with up to 128,000 output tokens — the published specification for Claude Mythos 5.1 and Claude Fable 5.1, released by Anthropic on September 1, 2026.

Claude Mythos is 📝Anthropic's Mythos-class model tier. Every publicly released Mythos-class model — 📝Claude Fable 5 and 📝Claude Mythos 5 in June 2026, and their 5.1 successors in September 2026 — shares the same 1M-token window. Anthropic bills the entire window at the standard per-token rate, with no long-context surcharge. Claude Mythos Preview, the restricted research model behind Project Glasswing, belongs to the same family, but Anthropic has not published its context window.

Key Facts

  • Context window: 1,000,000 input tokens (default and maximum)
  • Max output: 128,000 tokens on the synchronous Messages API
  • Capacity: about 555,000 English words, or 2.5 million characters, on the tokenizer introduced with Claude Opus 4.7
  • Current model IDs: claude-fable-5-1 (generally available); claude-mythos-5-1 (restricted to vetted organizations)
  • Pricing: $10 per million input tokens, $50 per million output tokens, $0.25 per million cached-read tokens
  • Knowledge cutoff: June 2026
  • Status as of: September 28, 2026

Sources: Anthropic — Claude Mythos 5.1 overview, Anthropic — What's new in Claude Fable 5.1.

How It Compares

Context size no longer separates Mythos-class models from the 📝rest of the 📝Claude lineup; reasoning depth, safeguards, and price do. Published specifications as of September 28, 2026:

  • Claude Fable 5.1 / Claude Mythos 5.1: 1M context, 128K output, $10 / $50 per MTok
  • 📝Claude Opus 5.5: 1M context, 128K output, $4 / $20 per MTok (released September 22, 2026)
  • Claude Sonnet 5.5: 1M context, 128K output, $2 / $10 per MTok
  • Claude Haiku 4.5: 200K context, 64K output, $1 / $5 per MTok

Anthropic describes Claude Opus 5.5 as performing at the level of Claude Fable 5.1 on most work at lower cost, which makes Opus 5.5 the default choice for long-context workloads that do not need Mythos-class reasoning. Source: Anthropic — Introducing Claude Opus 5.5.

What Fits in 1 Million Tokens

One million tokens holds roughly 555,000 words on the current Claude tokenizer, which produces about 30% more tokens for the same text than models before Claude Opus 4.7. Older figures of "750,000 words per million tokens" describe the earlier tokenizer and overstate current capacity.

In practice, the window can hold an entire mid-sized codebase, a large contract set, or hundreds of research papers in a single request. For agentic work, it holds a project's source alongside the model's reasoning, tool outputs, and intermediate results in one session.

Claude Fable 5.1 improved long-context reasoning over Claude Fable 5, connecting details across the full window more reliably, and cut cached-read prices to 2.5% of base input. Long sessions that re-read a cached prefix now cost a quarter of the Fable 5 rate.

Project Glasswing and Mythos Preview

Claude Mythos Preview is the restricted frontier model Anthropic deployed through 📝Project Glasswing, its initiative to secure critical software. Launch partners were Amazon Web Services, Anthropic, Apple, Broadcom, Cisco, CrowdStrike, 📝Google, JPMorganChase, the Linux Foundation, Microsoft, NVIDIA, and Palo Alto Networks, plus more than 40 additional critical-infrastructure organizations, backed by $100M in usage credits. Source: Anthropic — Project Glasswing.

Holding whole projects in context during agentic sessions, Mythos Preview produced 181 working exploits against Firefox's JavaScript engine (versus 2 for Claude Opus 4.6), 595 tier 1–2 crashes in OSS-Fuzz evaluation, and 89% severity agreement with expert reviewers across 198 vulnerability reports. Source: Anthropic — Claude Mythos Preview.

Reported, Not Confirmed

These claims circulate widely but are not confirmed by Anthropic:

  • Mythos Preview pricing of $25 / $125 per MTok. [reported]
  • Mythos Preview context window, model ID, knowledge cutoff, and an April 7, 2026 launch date. [reported]
  • Leaked benchmark figures such as 93.9% SWE-bench Verified. [rumor]
  • Parameter count and training cost estimates, including "~10 trillion parameters." [rumor]

Rumor sources: The Decoder, AI & Analytics Diaries.

FAQ

How big is the Claude Mythos context window?

The Claude Mythos context window is 1 million input tokens with a 128,000-token maximum output. This applies to Claude Mythos 5.1, Claude Fable 5.1, and their June 2026 predecessors.

How many words fit in the Claude Mythos context window?

About 555,000 English words, or 2.5 million characters. The current Claude tokenizer produces roughly 30% more tokens per word than models before Claude Opus 4.7.

Did Claude Mythos 5.1 increase the context window?

No. Claude Mythos 5.1 and Claude Fable 5.1 keep the 1M-token window and 128K output limit of Claude Mythos 5 and Claude Fable 5. The 5.1 release improved long-context reasoning and cut cached-read prices by 75%.

Does Claude Mythos charge extra for long context?

No. Anthropic bills the full 1M-token window at the standard rate of $10 per million input tokens and $50 per million output tokens.

What is the Claude Mythos knowledge cutoff?

Claude Fable 5.1 and Claude Mythos 5.1 have a reliable knowledge cutoff of June 2026. Anthropic has not published a knowledge cutoff for Claude Mythos Preview.

What is the Claude Mythos Preview context window?

Anthropic has not published it. Only the generally released Mythos-class models — Fable 5, Mythos 5, Fable 5.1, and Mythos 5.1 — have confirmed 1M-token windows.

Does Claude Opus 5.5 have the same context window as Claude Mythos?

Yes. Claude Opus 5.5 also has a 1M-token context window and 128K max output, at $4 / $20 per million tokens.

Related

We think the headline was never the number. A million tokens is now standard across the Claude lineup; what Mythos-class reasoning adds is the ability to use all of it coherently. That is what makes whole-codebase audits and long agentic sessions practical, and it is the same property that led Anthropic to gate the least-restricted version behind vetting.

Contexts

Created with 💜 by One Inc | Copyright 2026