Tweet
Key Differences
- Claude Fable 5 (public/general version):
- The "safe for general use" version available to paid Claude users (Pro/Max plans, API, etc.).
- It has additional safety cl***ifiers/guardrails that detect and route sensitive queries (especially in cybersecurity, biology, chemistry, and model distillation) to the weaker Claude Opus 4.8 model.
- This makes it less capable on high-risk topics but fully usable for most tasks.
- First publicly available Mythos-cl*** model.
- Claude Mythos 5 (restricted version):
- The full-power version with safeguards lifted in key areas (particularly cybersecurity).
- Available only to a small group of vetted partners via Project Gl***wing (cyber defenders, critical infrastructure providers, and select biology researchers), in collaboration with the US government.
- Successor to the earlier Claude Mythos Preview.
Both are described as a major leap in performance:
- State-of-the-art (or near-top) on nearly all major benchmarks, especially coding/software engineering, long-horizon agentic reasoning (autonomous work over extended periods), knowledge work, vision, and scientific tasks.
- Excel at complex, multi-step tasks where their advantage grows with task length/complexity.
- 1 million token context window.
- Strong autonomous capabilities (e.g., handling large codebases, research, etc.).
- Pricing: $10 per million input tokens / $50 per million output tokens (roughly 2x the price of previous Opus models).
- Fable 5 was initially included in paid plans (until ~June 22, 2026), then shifted to pay-as-you-go credits.
- Note: Access to both models was reportedly suspended shortly after launch due to export control/national security directives.
In short, Fable 5 is the powerful but guarded public model you can (or could) use today, while Mythos 5 is the unrestricted version for trusted high-security use cases. They represent Anthropic's approach to responsibly releasing extremely capable models. For the official details, check Anthropic's announcement.