Mythos 5 (Claude Mythos 5)
Anthropic · Released Jun 2026
The strongest cybersecurity model in the world (83.8% CyberGym) — built on Fable 5's weights with safety classifiers lifted. Access was suspended after the Jun 12 export directive; partially restored Jun 26 to ~100 designated US critical-infrastructure organizations (Project Glasswing). Verdict stays CONDITIONAL: capability is elite, but access is gated by federal designation. Not a public API — usable only if your organization is on the Commerce Department list.
Is it right for you?
Good for
- Cybersecurity vulnerability discovery and exploit reproduction — 83.8% on CyberGym, agentic hacking across reconnaissance, discovery, and lateral movement
- Protein design and drug discovery — accelerated design process ~10x vs human operators in internal Anthropic studies
- Novel scientific hypothesis generation — preferred ~80% over Opus-class models by scientists in blinded head-to-head molecular biology comparisons
- Autonomous multi-day research — conducted novel genomics research over 1+ weeks with only high-level human input
Not good for
- General-purpose chat or development — restricted to Project Glasswing partners only, not a public API
- Budget or high-volume workloads — $50/1M output is premium pricing, and access is capped by invitation
- Users outside US/approved geographies — export control ban blocks all foreign national access, inside or outside the US
Pricing
Input
$10 / 1M tokens
Output
$50 / 1M tokens
Context
1M tokens (restricted access)
Benchmarks
| Benchmark | Score | Source |
|---|---|---|
| CyberGym | 83.8% | Source |
Verdict history
Sources
- Anthropic — Fable 5 and Mythos 5 launch announcementJun 2026
- Anthropic — Statement on US government directiveJun 2026
- WIRED — OpenAI Launches Full-Scale Effort to Patch Open-Source BugsJun 2026
- Axios: White House quiet on OpenAI's Mythos-like modelJun 2026
- CloudZero — Claude Mythos pricing 2026Jun 2026
- ClaudeFa.st — Fable 5 & Mythos 5 benchmarksJun 2026
Verification log
- Pricing— No changes
Automated agent
- Pricing— No changes
Automated agent
- Status— Updated
Automated agent
Summary corrected: removed stale 'currently inaccessible / no timeline for restoration' language; now states restricted-access as of Jul 7 — restored Jun 26 to 100+ designated US critical-infrastructure orgs (incl. non-American employees) via the Lutnick authorization, first off-ramp from the June 12 export ban. Verdict stays conditional (access gated by federal designation).
- Pricing— No changes
Automated agent
- Pricing— No changes
Automated agent
Pricing unchanged: $10/$50 per 1M tokens. Access partially restored Jun 26 to 100+ US orgs (PCMag). Verdict_summary now outdated (says 'inaccessible to all' but access partially restored).
- Profile— No changes
Imported at launch
- Pricing— No changes
Imported at launch