Anthropic announced Fable, the first publicly available model in a new series that includes many capabilities from its advanced Mythos system. The debut comes shortly after the company said it was accelerating plans to release a Mythos-capable model with extensive safeguards.
Mythos was first introduced in early April through Project Glasswing, a limited-access program that allowed a small group of partners, including Apple and NVIDIA, to test Anthropic’s most advanced AI for strengthening defenses against AI-driven cyber threats. The initiative also attracted attention in Washington, contributing to renewed discussions around AI regulation.
With Fable, Anthropic is making a broader set of Mythos-level capabilities available to the public. The model carries the version five designation in line with the company’s overall numbering system and is described as its strongest public release so far. According to Anthropic, Fable achieved better results in internal testing than Opus 4.8, OpenAI’s GPT-5.5, and Google’s Gemini 3.1 Pro, particularly on longer and more demanding tasks.
According to Anthropic, Fable is particularly good at coding and working through large documents. It also takes a big step forward in vision tasks, which used to be one of the company’s weaker areas. While Claude 3.7 Sonnet struggled to understand Pokémon Red without extra visual assistance, Fable can make its way through Pokémon FireRed with very little help. Anthropic says that same capability allows the model to pull detailed information from scientific graphics and rebuild web apps directly from screenshots.
To balance safety with a rapid rollout, Anthropic added a cautious routing system to Fable. Under certain circumstances, Claude will redirect prompts to the less powerful Opus 4.8 model instead. The company said the safeguards were intentionally configured conservatively, meaning some harmless requests may be flagged, though the system activates in fewer than 5 percent of sessions on average. Anthropic added that it is working to reduce unnecessary triggers as more advanced models are introduced.
In addition to Fable, Anthropic has launched Mythos 5, which is powered by the same base model but operates with fewer safeguards. The model will initially be available through Project Glasswing before becoming accessible to a broader audience through a trusted access initiative.
Fable 5 will be available to Claude subscribers at no extra charge until June 22. Afterward, access will be charged at a rate of $10 per million input tokens and $50 per million output tokens. According to Anthropic, the goal is to make the model part of its standard subscription offerings again once infrastructure capacity permits.
Maybe you would like other interesting articles?