Unknown · v5 · 2× · vu le 08 août 2026
Claude Mythos 5 is a Frontier language model released by Anthropic on June 9, 2026, belonging to the newly introduced "Mythos Class," positioned above Opus. It shares weights with the publicly available Claude Fable 5 but omits its security classifiers for cybersecurity and biology, and is accessible exclusively to vetted partners through "Project Glasswing" (cyber defenders, infrastructure operators, biomedical researchers). The model supports a 1-million-Token context window, text and image input, and advanced Thinking, achieving top scores on SWE-bench Verified (95.0%) and the cybersecurity-specific ExploitBench (78.0%). Due to risks in cyber-offensive capabilities, access was temporarily subject to U.S. export restrictions.
Fonctionnalités
| Key Benchmark (%) | 95.0% on SWE-bench Verified; 78.0% on ExploitBench (vs. 40.0% Opus 4.8) |
| Context Window (Tokens) | 1,000,000 input tokens / 128,000 output tokens |
| License | Proprietary, closed model, no open weight download |
| Multimodality | Text and image input, text output |
| Platform | Cloud-only via Anthropic API / Claude Platform; access restricted to vetted partners via Project Glasswing |
| Price per 1M Tokens | $10 input / $50 output per 1M tokens |
| Release Date | June 9, 2026 (predecessor "Mythos Preview" unveiled April 7, 2026) |