Model Details
Claude Mythos Preview
ACCESS SUSPENDED: As of June 2026, access to Claude Mythos Preview has been suspended under a US government export control directive. See Anthropic's statement for details. Teams seeking equivalent capabilities should evaluate Claude Opus 4 or Claude Fable 5 if/when access is restored.
At a Glance
Benchmark highlights (prior to suspension)
- 93.9% SWE-bench Verified — near-top software engineering performance
- 83.1% CyberGym — highest CyberGym score on Benchgen
- Advanced reasoning and specialised deployment capabilities
- Multimodal vision support
Status
- Access SUSPENDED under US government export control directive (June 2026)
- Previously restricted to approved API access
- Specialised model line beyond standard Claude tiers
What Claude Mythos Preview Is
Claude Mythos Preview is part of Anthropic's Mythos model line — a specialised series of advanced models positioned beyond the standard Claude Sonnet/Opus tiers. The Mythos Preview was released in early 2026 as a preview of Anthropic's most advanced capabilities, targeting specialised research and enterprise deployment scenarios.
Its benchmark profile is striking: 93.9% SWE-bench Verified (trailing only Claude Fable 5's 95%) and 83.1% CyberGym (the highest CyberGym score in the Benchgen dataset) indicate a model positioned at the frontier of software engineering and cybersecurity capability.
In June 2026, a US government export control directive suspended access to both Claude Fable 5 and Claude Mythos (the Preview and any production versions). Anthropic's statement notes that this applies to US and international customers. Teams planning deployments that require this capability level should monitor Anthropic's communications for access restoration.
Access Status
Current status: SUSPENDED (June 2026)
Per Anthropic's official statement:
- Access to Claude Mythos Preview has been suspended under a US government export control directive
- This applies to both US and international customers
- Anthropic is working with government partners on the path forward
Recommended alternatives while suspended:
- Claude Fable 5 (also suspended — same directive)
- Claude Opus 4 (Thinking) — available, strong reasoning capability
- Gemini 2.5 Pro — 21.64% HLE, 73.6% LiveCodeBench
Specifications
| Field | Value |
|---|---|
| Organization | Anthropic |
| Parameters | Undisclosed |
| Context window | Undisclosed |
| License | Proprietary (API only) |
| Release date | January 2026 (Preview) |
| Knowledge cutoff | October 2025 |
| Modality | Text + Vision (multimodal) |
| Current access | SUSPENDED |
Public Benchmark Scores
| Benchmark | Score | Source | Date |
|---|---|---|---|
| SWE-bench Verified | 93.9% | Benchgen evaluation | 2026-06 |
| CyberGym | 83.1% | Benchgen evaluation | 2026-06 |
Claude Mythos Preview vs Alternatives
| Model | SWE-bench Verified | CyberGym | Access |
|---|---|---|---|
| Claude Mythos Preview | 93.9% | 83.1% | SUSPENDED |
| Claude Fable 5 | 95.0% | — | SUSPENDED |
| Claude Opus 4 (Thinking) | — | — | Available |
| Gemini 2.5 Pro | — | — | Available |
Claude Mythos Preview's 83.1% CyberGym score is exceptional — no other model tracked on Benchgen approaches this score. Its 93.9% SWE-bench Verified is second only to Claude Fable 5 (95%). Both models are currently inaccessible.
Related Models & Benchmarks
- Compare: Claude Fable 5 · Claude Opus 4 (Thinking) · Gemini 2.5 Pro
- Benchmarks: SWE-bench Verified · CyberGym
Frequently Asked Questions
What is Claude Mythos Preview?
Claude Mythos Preview is part of Anthropic's advanced Mythos model line, scoring 93.9% SWE-bench Verified and 83.1% CyberGym. Access has been suspended since June 2026 under a US government export control directive.Can I still access Claude Mythos Preview?
No. As of June 2026, access to Claude Mythos Preview has been suspended under a US government export control directive. Monitor Anthropic's official communications for updates.What should I use instead of Claude Mythos?
For software engineering tasks, Claude Opus 4 (Thinking) is the best currently-available Anthropic option. For maximum coding/reasoning performance across available models, consider Gemini 2.5 Pro or o4-mini-high.Benchmark scores from Benchgen evaluations conducted prior to access suspension. Access status per Anthropic's official statement (June 2026). Last updated 2026-07-24.