Anthropic launches Fable 5, a safety-limited Mythos-class model

Anthropic has launched two models under the Mythos umbrella: Claude Fable 5, which is broadly accessible, and Mythos 5, which is limited to participants in what Anthropic calls "project glass wing" — likely heavy-spending enterprise customers with specific use cases. Sam Witteveen characterizes Fable 5 as a rebranded, safety-constrained version of the Mythos 5 capability, noting that Anthropic describes it as a "Mythos class model" made safe for general use and claims its capabilities exceed those of any model they have previously made generally available.

On pricing, both Fable 5 and Mythos 5 are priced at double the cost of Opus 4.8, which Witteveen notes is lower than he anticipated — he had expected a quadruple price increase.

On pricing, both Fable 5 and Mythos 5 are priced at double the cost of Opus 4.8, which Witteveen notes is lower than he anticipated — he had expected a quadruple price increase. Anthropic attributes the more accessible pricing in part to an inference compute deal, and explicitly notes the new pricing is less than half that of the Claude Mythos preview. Benchmark results show strong performance over Opus 4.8 and GPT 5.5, with a particularly large jump on SWEbench Pro and Frontier Code, a new benchmark from Ignition, where Fable 5 scores more than double Opus 4.8's 13.4. However, Witteveen notes the gains are less dramatic on tool use and computer use benchmarks, and cautions that real-world assessment — especially for agentic coding tasks — will require days or weeks of testing. The video also previews safety restrictions on the model, including a biology-related trigger test, and notes upcoming changes around data retention and subscriptions; the source transcript is truncated before those details are fully covered.

Key facts

1. 01Anthropic released two models: Claude Fable 5 (broadly available) and Mythos 5 (restricted to 'project glass wing' participants).
2. 02Fable 5 is described as a 'Mythos class model' made safe for general use, with capabilities claimed to exceed any previously generally available Anthropic model.
3. 03Pricing for both Fable 5 and Mythos 5 is double that of Opus 4.8, but less than half the price of the Claude Mythos preview.
4. 04On the Frontier Code benchmark from Ignition, Fable 5 scores more than double Opus 4.8's score of 13.4.
5. 05Fable 5 also outperforms GPT 5.5 on benchmarks, though gains on tool use and computer use benchmarks are described as less substantial.
6. 06The video covers safety restrictions on the model, including a biology-related trigger test, and notes upcoming data retention and subscription changes.

Topics