Anthropic launches Claude Fable 5.2 and Mythos 5.2
Anthropic has launched Claude Fable 5.2 and Mythos 5.2 with claimed gains in coding, reasoning and vision, plus new pricing and access rules.
Anthropic has launched Claude Fable 5.2 and Claude Mythos 5.2, two versions of the same underlying model aimed at demanding technical work. The company says the models improve coding, reasoning and vision, while Fable is broadly available and Mythos is restricted to organisations applying through Anthropic’s cybersecurity and life-sciences verification programmes. 1
Two versions, different access
Anthropic describes Fable as the general-access version, with safeguards for cybersecurity and biology work. Mythos is offered through the company’s Cyber Verification and Life Sciences Verification programmes for organisations pursuing those applications. Anthropic says Fable 5.2 is available across its platforms, including Amazon Web Services, Google Cloud and Microsoft Azure. 1
The company says both models communicate more clearly than their 5.1 predecessors and generate outputs up to 30% faster. Anthropic also says the models cost $8 per million input tokens and $40 per million output tokens. An optional fast mode doubles those prices to $16 per million input tokens and $80 per million output tokens, while aiming to increase token generation speed by up to 2.5 times. 1
Anthropic’s benchmark claims
Anthropic reports gains across coding, scientific research, computer use, robotics, visual chart recognition and other evaluations. Its announcement says Fable 5.2 reaches 49.1% on FrontierCode 2.0 at Xhigh effort, compared with 32.4% for Fable 5.1 and 45.5% for Opus 5.5. It also reports 69.9% on FrontierCode SWE v2 and 89.3% on OSWorld 2.1’s offline partial set. 1
Those figures are Anthropic’s own testing results. The company links a system card with methodology and safety details, but no independent benchmark study was identified for this launch. The results are therefore useful as a description of Anthropic’s positioning, not as a settled industry-wide performance verdict.
What the models are for
Anthropic says Fable and Mythos are designed for open-ended technical challenges, including complex software migrations, scientific reasoning and agent orchestration. The company recommends Claude Opus 5.5 as a daily-driver starting point for most coding and knowledge tasks, with Fable or Mythos reserved for work that needs more capability. Fable can also act as an adviser or orchestrator for smaller models, according to the announcement. 1
Anthropic says Enterprise Frontier Safeguards is entering private beta with selected customers. It also says prerelease safety testing found the models performed as well as or better than recent models on most of its alignment measures. Those are company assessments, and the linked system card is the appropriate place to examine the methodology and limitations.
Our opinion
The notable part of this launch is the split between capability and access. Anthropic is presenting Fable 5.2 as a costly specialist for difficult technical work, while Mythos adds a controlled route for high-risk research areas rather than becoming a standard public endpoint. That may be sensible governance, but it also means the headline benchmark claims will initially be hard for outsiders to reproduce. The pricing is lower than the previous models according to Anthropic, yet the fast mode quickly becomes expensive at scale. This is a significant model release, but readers should wait for independent testing before treating the benchmark table as a final ranking.
For developers, the immediate news is availability and pricing. For everyone else, the more consequential question is whether the claimed gains survive testing outside Anthropic’s own evaluation suite.