Claude Mythos 5 is Anthropic’s most capable model to date — and it is not generally available. It ranks #1 overall on BenchLM (83.1/100) and #1 for coding among 135 models, but access is restricted to a small group of vetted government cyber-defense partners and select research organizations. Here is everything that is public about it as of August 2026.
Benchmarks: What’s Confirmed
Mythos 5 holds the top BenchLM public score at 83.1, ahead of every other model including GPT-5.6 Sol and Claude Fable 5. Its strongest categories are Coding (#1 of 135) and Agentic tasks (#3 of 129). Anthropic has not published full independent benchmark suites, so treat broad reasoning/math claims carefully — much of the evidence comes from restricted partners.
Pricing: $10/$50
The published API price is $10 per million input tokens and $50 per million output tokens, with cached input at $1 and batch pricing at $5/$25. That is roughly double Claude Opus 5 and double GPT-5.6 Sol on input — the most expensive widely-published frontier rate today.
Mythos 5 vs Fable 5: Same Model, Different Access
Claude Fable 5 is the general-purpose release of the same underlying model with safety classifiers layered on. Queries that match cybersecurity, biology, or distillation rules are automatically rerouted to Claude Opus 4.8. Mythos 5 removes those classifiers for approved partners. Both are billed identically at $10/$50.
Why It Matters for Buyers
Even if you cannot access Mythos 5, its existence reshapes tiering. Mid-tier models like Sonnet 5 and Opus 5 now anchor the “production frontier,” while frontier-class (Mythos-level) capability sits behind access controls. Watch for a gradual expansion of trusted access — Anthropic has said biology researchers are next in line.
What to Use Instead
For production, Claude Opus 5 ($5/$25) delivers most Mythos-level capability without access restrictions, and Fable 5 is available via the Claude API for teams that need the top of the stack with classifiers applied. See our AI Model Comparison 2026 for positioning.
