Claude Fable and Claude Mythos 5.1: One Model, Two Gates
Anthropic has shipped version 5.1 of its frontier model under two names, Claude Fable and Claude Mythos, and the split tells you more about how powerful AI gets deployed responsibly than any single benchmark score does.
Anthropic has shipped version 5.1 of its frontier model under two names, Claude Fable and Claude Mythos, and the split tells you more about how powerful AI gets deployed responsibly than any single benchmark score does.
Same weights, different doors
Anthropic states plainly that Claude Fable 5.1 and Claude Mythos 5.1 are the same model with different safeguards. Fable is generally available now, reachable through the Claude API as claude-fable-5-1, and through Amazon Web Services, Google Cloud, and Microsoft Azure. Mythos, by contrast, sits behind trusted access programs, including one partnered with a US government body for its advanced biology capabilities. If you're building a product today, Fable is the version you'll actually touch. Mythos exists as a separate gate on the same underlying system, reserved for a narrower audience. That distinction becomes more concrete once you look at what each side is being asked to do.
What the safeguards actually change
The Enterprise Frontier Safeguards store customer data in customer-controlled cloud infrastructure — that is the specific claim Anthropic makes, not a broader guarantee about how data moves during use. On the security side, Claude Fable 5.1's safeguards produce about 60% fewer cybersecurity interventions than the previous safeguards on Fable 5. Fable can identify software vulnerabilities directly, and Mythos 5.1 powers Claude Security, a vulnerability-scanning product built on the same weights. Together, these facts describe a model being placed inside security-sensitive pipelines, with the safeguard layer as the mechanism Anthropic points to for making that placement viable.
The numbers behind the API call
Pricing dropped in real terms: Claude Fable 5.1 runs at $10 per million input tokens, $50 per million output tokens, and $0.25 per million tokens for cache reads. Anthropic reports that typical token-billed workloads cost about 25% less with 5.1 than with Fable 5, and highly agentic workloads can cost up to about 45% less. On benchmarks, Fable 5.1 scored 55.8% on Terminal-Bench 4.0, 52.6% on Terminal-Bench-Science 0.1, 1853 on GDPval-AA v2, 77.9% partial and 41.7% strict on OSWorld 2.0, 60.9% without tools and 65.0% with tools on Humanity's Last Exam, 31.4% on AutomationBench, and 73.4% on CursorBench 3.2.0. Anthropic does not characterize these benchmarks beyond the scores themselves, but the combination of lower cost and this spread of results is what a team would weigh when deciding whether a workload is now cheap enough to run continuously rather than sparingly.
What Mythos is doing while Fable is public
Mythos 5.1's restricted access hasn't kept it idle. It achieved a nearly 50% binder hit rate across 12 targets, sped up seven open-source deep learning models by up to 2.5 times, and reduced estimated GPU costs by 30% to 60% for optimized analyses. Fable, meanwhile, produced an elevation map covering one-third of Venus. The pattern across both names is the same: concrete scientific and infrastructure output, run under whichever safeguard tier matches the sensitivity of the task at hand.
Where this leaves someone deciding what to build
If your work involves agentic coding, vulnerability scanning, or long-running automation, Fable 5.1 is already available through the Claude API and the three major clouds at published prices, so the cost figures above are something you can test against your own workload directly. If your work touches biosecurity research or another domain gated by trusted access, Mythos 5.1 is the program to inquire about rather than an API you'd casually call. Sorting out which tier fits a given use case, which safeguard model matches a compliance posture, or how the new pricing changes an existing agent pipeline's economics is exactly the kind of scoping work worth doing with people who build on these systems regularly, rather than guessing from a pricing page alone.
Specs
- Typical workload cost reduction: About 25% versus Claude Fable 5
- Highly agentic workload cost reduction: Up to about 45% versus Claude Fable 5
- Cache read price: $0.25 per million tokens
- Input token price: $10 per million tokens
- Output token price: $50 per million tokens
- Model relationship: Claude Fable 5.1 and Claude Mythos 5.1 are the same model
- Fable availability: Generally available
- Mythos availability: Trusted access programs
Sources
- Anthropic — Anthropic