claude-mythos-5-1 — pricing, context window and limits
The same model as Fable 5.1 — same prices, limits and API behaviour — offered only to approved Project Glasswing organisations. Unlike Mythos 5 it runs safeguards, so handle the refusal stop reason; unlike Fable 5.1 it skips the history-editing check. Not offered on Claude Platform on AWS.
Price per million tokens
| Input | $10 |
|---|---|
| Output | $50 |
| Cache write, 5-minute TTL | $12.50 |
| Cache write, 1-hour TTL | $20 |
| Cache read | $0.25 (0.025x input) |
| Batch input / output | $5 / $25 |
Limits and behaviour
| Status | Current |
|---|---|
| Context window | 1M tokens |
| Max output | 128K tokens |
| Min cacheable prefix | 512 tokens |
| Effort levels | low, medium, high, xhigh, max |
| Effort default | high |
| Thinking | Always on — cannot be disabled |
| Forced tool_choice | Rejected (400) |
| Platforms | Claude API, Amazon Bedrock, Google Cloud, Microsoft Foundry |
| Access | Project Glasswing participants only (invitation) |
| Released | 1 Sept 2026 |
Lifecycle
| API model name | claude-mythos-5-1 |
|---|---|
| Retires no sooner than | 1 Sept 2027 |
Fable 5.1, for Project Glasswing organisations
Mythos 5.1 is the same model as Fable 5.1 — the same capabilities, limits and per-token prices, including the deep cache-read discount — offered only to organisations approved into Project Glasswing. If your organisation isn't in the programme, Fable 5.1 is the model you want, and nothing about it is a lesser version of this one.
The two ways its behaviour differs
First, from its predecessor: Mythos 5 ran no safety classifiers, so a refusal stop reason never
occurred on it. Mythos 5.1 runs safeguards that depend on the access programme an organisation was
approved under. Code migrating from Mythos 5 needs a branch for stop_reason: "refusal" before it
reads the response content, and a fallback plan for declined requests, where it previously needed
neither.
Second, from Fable 5.1: Mythos 5.1 does not run the history-editing check that makes Fable 5.1 reject replayed thinking blocks after an earlier turn changes. The model-binding rule still applies — Mythos 5.1 and Fable 5.1 read each other's thinking blocks, and Mythos 5.1 also reads Mythos 5's, while other models read neither. Editing history still restarts the prompt cache, so an append-only harness is still the cheaper one.
Where it is offered
The Claude API, Google Cloud, Microsoft Foundry, and Amazon Bedrock in a limited, unlisted form — but not Claude Platform on AWS, unlike Fable 5.1. Whether an organisation's Mythos 5 access carries over automatically was left open at launch, so confirm with the account team before switching ids.
Related
See claude-fable-5-1 for the generally available model, and handling refusals without an outage for the refusal handling this model now needs.
Verified 2026-09-30 against claude-api skill — shared/models.md, model-migration.md + SKILL.md model table and Anthropic's pricing and model-deprecation pages (canonical: https://platform.claude.com/docs/en/about-claude/models/overview, https://platform.claude.com/docs/en/about-claude/pricing, https://platform.claude.com/docs/en/about-claude/model-deprecations).Could not confirm: Its Amazon Bedrock listing is described as us-east-1 only and not publicly listed, and whether Mythos 5 access carries over automatically was open at launch.
Verified 2026-09-30 against Anthropic's model-deprecations status table and pricing page (canonical: https://platform.claude.com/docs/en/about-claude/model-deprecations, https://platform.claude.com/docs/en/about-claude/pricing).Could not confirm: These dates apply to the Claude API, Claude Platform on AWS and Microsoft Foundry. Amazon Bedrock and Google Cloud keep their own retirement schedules, which we do not track.