Quick answer
Claude Mythos 5 — Anthropic's most powerful model to date — has been re-enabled after a brief suspension. Available to 100+ trusted US organisations. Strong on cybersecurity, autonomous coding, and enterprise AI agents. Mythos 5 sits above Opus 4.8 in the Anthropic hierarchy and competes directly with GPT-5.6 Sol.
Anthropic released Claude Mythos 5 several months ago. It was suspended temporarily following the company's safety review framework triggering on certain capability evaluations. After remediation, the model is back online — but only for a tightly controlled set of trusted partners. Here's what we know.
What Mythos 5 is
- Anthropic's top-of-stack model — above Opus 4.8 in capability tiers
- ~95% reliability on agent benchmarks, leading SWE-Bench scores
- Designed for autonomous, multi-hour task execution
- Strongest cybersecurity-task performance of any frontier model
- Native multi-tool orchestration optimised for production agent workflows
Why it was suspended
Anthropic has a published responsible scaling policy with explicit capability thresholds. When Mythos 5 hit a threshold on certain biosecurity-related evaluations, the policy triggered. Anthropic paused commercial access while they hardened safeguards — additional constitutional training, refined refusal patterns, expanded red-teaming. The framework worked as designed. The pause was the system being responsible, not the model being broken.
Who can use it
- Approximately 100 trusted US organisations: Fortune 500 enterprises, US government contractors, named research labs
- Access is contract-based, not self-serve
- No general API availability today
- EU/UK access is going through separate regulatory review
- Public availability timing unclear
What's it good at
- Cybersecurity work — penetration testing assistance, malware analysis, vulnerability research (in controlled, authorised contexts)
- Autonomous coding — multi-hour engineering tasks, complex refactors, full feature builds
- Enterprise AI agents — multi-step orchestrated workflows with high reliability
- Long-context analysis — full codebases, multi-document research, deep legal review
If you're evaluating Anthropic for an enterprise contract right now, Mythos 5 access is becoming a quiet differentiator in procurement conversations. Sales reps can put it on the table — but only after security and compliance reviews.
What it means for the AI landscape
Mythos 5 and GPT-5.6 Sol are both gated by capability-based access reviews. This is the new shape of frontier AI: highest-tier capabilities are no longer "buy with a credit card" products. They're enterprise-procurement products with safety and compliance gates. That's likely the new normal for the most powerful tier of models from here on.
Bottom line
Mythos 5 is back. It's the most powerful Claude. It's also the most gated. Frontier AI is splitting into "broadly available capable" (Opus 4.8, GPT-5) and "restricted top-tier" (Mythos 5, GPT-5.6 Sol). For most builders, Opus 4.8 remains the right default. For approved enterprises, Mythos 5 is the new bar.
