Anthropic has announced Claude Fable 5.1 and Claude Mythos 5.1 — the same underlying model with different levels of safeguards. Fable 5.1 is generally available, while Mythos 5.1 is currently limited to trusted access programs focused on cybersecurity and life sciences.
The central idea is not simply a better chatbot response. It is a model that can work through a complex task over a long period: research, write code, use tools, check its output, and correct mistakes.
More performance, with a focus on cost
According to Anthropic, Fable 5.1 scored 52.6% on Terminal-Bench-Science 0.1, compared with 24.7% for Fable 5. The company also reports gains across coding, reasoning, and computer-use tasks.
These figures come from Anthropic, so they need to be read in context. The scientific benchmark has a standard error of roughly 3.5–4.5 percentage points. That makes the result a useful signal, not proof that the model is better in every environment.
Anthropic estimates that Fable 5.1 will cost about 25% less for typical token-billed workloads, with savings of up to approximately 45% for highly agentic work. The change is partly related to lower cache-read pricing. Actual savings will depend on the workflow and how much context is reused.
Capability is not the same as permission
Fable 5.1 and Mythos 5.1 share the same base model, but their safeguards differ. Mythos is intended for more sensitive scenarios and remains restricted to a dedicated access program.
Anthropic also says Fable 5.1 can help discover software vulnerabilities, but is not intended to develop exploits. That distinction matters: stronger models do not remove the need for clear permissions, monitoring, and human review.
What is actually changing
The most important part of the announcement is not another benchmark table. Model competition is increasingly about completing a long chain of work reliably and at a cost that can be justified.
If a model can run for hours, verify its own work, and recover from errors without losing the thread — while using fewer resources — it starts to look less like an experiment and more like a practical work tool.
That future still depends on independent testing, explicit access controls, and human checkpoints.