claude-mythos-preview Is Deprecated — But Its Replacement Is Currently Unavailable

Anthropic has added claude-mythos-preview to its official model deprecations list, with a hard retirement date of June 30, 2026. After that date, API calls to the model will fail. The documented migration target is claude-mythos-5. There is, however, a complication: claude-mythos-5 is currently suspended and unavailable to all customers, following a US government export control directive issued on June 12, 2026.

So if you’re a developer with claude-mythos-preview in production, you need a plan, and that plan can’t simply be “swap the model string.”

What Anthropic’s deprecation notice actually says

Anthropic’s model deprecations page follows a consistent lifecycle: models move from Active to Legacy to Deprecated, and finally to Retired. A Deprecated model still works, but it has a replacement and a retirement date assigned. Once it hits Retired, requests return errors.

claude-mythos-preview is now in the Deprecated stage. The retirement date is June 30, 2026. The migration guide in the developer docs walks through the transition, and Anthropic has notified impacted customers by email as per its standard deprecation policy.

Under normal circumstances this would be a routine upgrade notice. claude-mythos-5 benchmarks meaningfully ahead of its predecessor — 80.3% on SWE-bench Pro, versus 77.8% for Mythos Preview and 69.2% for Opus 4.8 — and it launched at less than half the price: $10 per million input tokens and $50 per million output tokens. Routine, straightforward, migrate and move on.

The circumstances are not routine.

Why the migration target is currently unavailable

claude-mythos-5 launched on June 9, 2026 alongside claude-fable-5, and access was pulled three days later. Anthropic received a directive from the US Commerce Department, signed by Secretary Howard Lutnick, instructing the company that any export, re-export, or domestic transfer of these models to non-US persons required prior government approval.

Rather than attempt to selectively block access — which would have meant cutting off a substantial portion of its user base, including some of its own foreign-born staff — Anthropic disabled both models entirely for all customers. Every other Claude model remains available. It is specifically Fable 5 and Mythos 5 that are offline.

Anthropic has been direct about its disagreement with the order. The company’s public statement notes that the government’s stated concern appears to centre on a “narrow, non-universal jailbreak” — one that essentially involves asking the model to read a codebase and identify software vulnerabilities. Anthropic says it independently verified that other frontier models, including OpenAI’s GPT-5.5, are capable of the same task. It also notes that the models went through thousands of hours of red-teaming by the US government, the UK AI Security Institute, and third-party organisations before launch, with no universal jailbreak found.

The company’s position is that applying this standard broadly “would essentially halt all new model deployments for all frontier model providers.” No timeline for resolution has been given publicly.

What this means for you right now

If you’re using claude-mythos-preview in production, you have until June 30, 2026 before your calls start failing. Here is a realistic view of your options:

If you have access to the Mythos program: You are presumably already aware of the suspension. Watch the models overview page and Anthropic’s email communications for when claude-mythos-5 access is restored. When it is, the migration guide covers what changes in the API.

If you need a fallback now: Anthropic recommends Opus 4.8 or Sonnet 4.6 as the practical alternatives while Mythos 5 is offline. Neither matches the benchmark performance of Mythos 5, but both are fully available and stable.

If your use case was cybersecurity-specific: claude-mythos-preview was initially released through Project Glasswing precisely because of its capabilities in areas like zero-day analysis. Those capabilities are what prompted the government’s concern about Mythos 5. Opus 4.8 will cover most code reasoning tasks, but if your workflow depends on the elevated cybersecurity capabilities that Mythos-class models unlock, there is no equivalent available on the Anthropic API right now.

On timeline risk: June 30 is the firm retirement date for Mythos Preview. The suspension of Mythos 5 has no stated end date. Those two timelines may not align. Build your contingency around Opus 4.8 or Sonnet 4.6 for the interval, rather than assuming claude-mythos-5 will be back in time.

A broader point worth noting

This is the first time a US government export control directive has targeted a specific commercial AI model’s general availability — not a chip, not training data, but deployed inference on a live API. Anthropic’s response, cutting off all customers to ensure blanket compliance, is a preview of how this kind of regulatory action plays out operationally.

For developers building on frontier AI APIs, government intervention in model access is now a documented, real-world risk. That probably warrants thinking about model abstraction layers and fallback logic in your architecture, not just as good engineering practice, but as basic resilience planning.

The migration guide is live in the developer docs now. The Anthropic models overview page is the right place to track when claude-mythos-5 access is restored.