“Fable is from the Latin fabula, ‘that which is told,’ akin to the Greek mythos,” Anthropic explained in a footnote on 9 June 2026. The two names it launched that day, Claude Fable 5 and Claude Mythos 5, belong to one and the same model. What separates them is a set of filters.
Fable 5 is for everyone: the first model from Anthropic's new top tier, which the company calls Mythos class and places above Opus, to go on general sale. Mythos 5 is the less filtered sibling, at first only for the cyber defenders of Project Glasswing, the programme Anthropic set up in April around Claude Mythos Preview. Both cost USD 10 per million input tokens and USD 50 per million output tokens.
- What: Claude Fable 5 (public, with safety classifiers) and Claude Mythos 5 (the same model with safeguards lifted in some areas, restricted access).
- Price: USD 10 per million input tokens and USD 50 per million output tokens, double Opus 4.8 and under half of Mythos Preview.
- Why it matters: Anthropic's most capable model reaches the public, with filters that hand risky requests to an older model.
A tier above Opus
“Fable 5's capabilities exceed those of any model we've ever made generally available,” Anthropic wrote, adding that its lead over the company's other models grows with the length and complexity of a task. The examples are vendor reported. Stripe said the model carried out a migration across a 50 million line Ruby codebase in a day, work that would have taken a whole team more than two months by hand. Fable 5 also finished the game Pokémon FireRed from screenshots alone, where earlier Claude models needed an elaborate helper harness.
Both versions share the same specifications: a 1 million token context window, up to 128,000 output tokens per request, and reasoning that is always switched on. The raw chain of thought is never returned to developers, only a summary or nothing.
How the filters split one model
Fable 5 runs with classifiers, separate AI systems that watch for misuse in three areas: cybersecurity, biology and chemistry, and distillation, meaning attempts to copy the model's abilities to train a rival. When one fires, Claude Opus 4.8 answers instead and the user is told. Anthropic said it tuned the filters conservatively, so they will sometimes catch harmless requests, but that they trigger on average in less than 5 percent of sessions. For now, most biology and chemistry questions fall back to Opus 4.8. An external bug bounty found no universal jailbreak, a prompt that disables the safeguards across the board, in more than 1,000 hours of testing, although the company said the UK AI Security Institute had made progress towards one in a brief initial window.
Without safeguards, Fable 5’s capabilities in areas like cybersecurity could be misused to cause serious damage.Anthropic, launch announcement, 9 June 2026
Mythos 5 is “the same underlying model as Fable 5, but with the safeguards lifted in some areas.” Glasswing partners could upgrade to it from Mythos Preview on launch day, and Anthropic said it was being deployed “in collaboration with the US government”. It claims Mythos 5 “has the strongest cybersecurity capabilities of any model in the world.” A separate trusted access programme for biology, giving selected researchers Fable 5 without the biology and chemistry filters but with the cyber ones still on, was promised for the coming weeks.
One more rule covers both versions. Anthropic now requires 30 day retention of all traffic on Mythos class models, including through cloud partners; TechCrunch reported that this applies even to enterprises that had zero retention agreements, and the developer documentation says the models are not available under zero data retention unless Anthropic expressly authorises it. The company says the data will not be used to train models and will help it catch new jailbreaks and reduce false positives.
Price and rollout
At USD 10 and USD 50, Fable 5 costs twice as much as Opus 4.8 and less than half the USD 25 and USD 125 that Glasswing participants were quoted for Mythos Preview. It was fully available from launch on the Claude API and on consumption based Enterprise plans, and also through Amazon Bedrock, Google Cloud and Microsoft Foundry. For subscribers Anthropic moved more cautiously: Fable 5 was included at no extra cost on Pro, Max, Team and seat based Enterprise plans until 22 June, and from 23 June it required usage credits; Anthropic said it aimed to make it a standard part of the plans again once capacity allowed.
What the safety report says
The system card treats the model as CB-1, Anthropic's level for help with known chemical and biological weapons, and applies ASL-3 protections to Fable 5. Anthropic concludes it does not cross the higher CB-2 threshold for novel weapons, but calls this “a much less clear judgment than for previous models” and writes that “the unsafeguarded Mythos 5 can significantly uplift well-resourced threat actors.” It judged overall alignment risk as very low, with test behaviour roughly comparable to Opus 4.8.
What happened next
On 12 June 2026 the US government applied export controls to both models, and because Anthropic had no reliable way to check users' nationality in real time, it suspended them for everyone. The controls were lifted on 30 June: Fable 5 returned globally on Anthropic's own platforms on 1 July, with the cloud partners to follow, and Mythos 5 came back for a set of US organisations after government approval on 26 June. On 1 September 2026 Anthropic followed up with Claude Fable 5.1 and Claude Mythos 5.1.
Comments
Comments are reviewed before they appear.
No comments yet. Be the first to share your thoughts.