In an announcement published September 29, Anthropic said Claude Fable 5.1 was generally available on its listed platforms, while Claude Mythos 5.1 was limited to trusted-access programs. The company describes both as the same underlying model, with Mythos configured with more permissive safeguards for vetted users.
One underlying model, two access paths
| Model | Underlying model and safeguards | Access described by Anthropic |
| Claude Fable 5.1 | Same underlying model; more restrictive safeguards than Mythos 5.1 | Generally available on Anthropic’s listed platforms |
| Claude Mythos 5.1 | Same underlying model; more permissive safeguards for vetted cybersecurity and life-sciences work | Trusted-access programs; initially offered to a set of U.S. organizations |
Anthropic says the distinction is about safeguards and access, not a different underlying model. Mythos is intended for vetted users whose cybersecurity or life-sciences work encounters restrictions under Fable’s safeguards.
What Anthropic reports about performance and research
Anthropic reports that Claude Fable 5.1 scored 55.8% on Terminal-Bench 4.0, compared with 42.0% for Claude Fable 5. On Terminal-Bench-Science 0.1, it reports 52.6% for Fable 5.1 and 24.7% for Fable 5; Anthropic gives the standard error for that test as ±3.5–4.5 points per model.
The company also reports 60.9% for Mythos 5.1 on Terminal-Bench 4.0, against 55.8% for Fable 5.1, and attributes the difference to safeguard interventions. These scores describe results on named evaluations.
Anthropic says Fable 5.1 used radar imagery from NASA’s Magellan mission and an earlier map to create an elevation map covering about one-third of Venus. The company reports 2–3 km detail, compared with 10–20 km in the earlier map, and height accuracy up to 25% higher.
For computational biology, Anthropic says Mythos 5.1 accelerated seven open-source deep-learning models by up to 2.5× while producing identical outputs. It estimates GPU-cost reductions of 30–60% for certain genome-scale analyses.
A screen-recorded Claude Code session shows Fable 5.1 generating a 3D animated bear riding a bicycle.
Safeguards shape what each model can do
Anthropic says Fable 5.1’s cybersecurity safeguards produce around 60% fewer interventions per Claude Code session than the safeguards used with Fable 5. The model can identify software vulnerabilities for defensive work; exploit development is restricted, while penetration testing and binary-based vulnerability scanning are redirected to other models.
For benign elementary-biology and medical requests, Anthropic reports 85% fewer safeguard triggers than under the safeguards first launched with Fable 5. Outside the Mythos program, life-sciences research and development requests are routed to Opus models.
Those limits help explain why Mythos has a separate access path. Anthropic describes the Cyber Verification Program as a route for vetted defensive-cybersecurity users and the Life Sciences Verification Program as a program for vetted professionals doing research and development. At the time of its announcement, Mythos access was initially limited to a set of U.S. organizations; Anthropic said it planned to expand to additional domestic and international partners.
API pricing and model limits
For developers, Anthropic lists the API model ID as claude-fable-5-1, with a one-million-token context window and a maximum output of 128,000 tokens. Its listed rates are $10 per million input tokens, $50 per million output tokens and $0.25 per million cache-read tokens.
Anthropic estimates that token-billed workloads cost about 25% less than with Fable 5 in typical use, and up to about 45% less for highly agentic work. The estimate draws on four weeks of default-effort usage in August 2026.
