Anthropic releases Claude Fable 5.2 and Mythos 5.2…

Fable and Mythos are the highest-positioned model series in Anthropic's product lineup, mainly targeting complex, open-ended technical tasks. By comparison, Haiku, Sonnet, and Opus handle everyday tasks of different scales and complexity. Anthropic recommends that ordinary programming and knowledge work can usually start with Claude Opus 5.5, and only when facing particularly complex, time-consuming tasks is it necessary to switch to Fable or Mythos.

In terms of performance, Anthropic says Fable 5.2 and Mythos 5.2 not only generate content faster but are also better at clearly expressing their reasoning when handling complex tasks. When completing difficult tasks at medium reasoning effort, Fable 5.2 can also use fewer tokens than Fable 5.1, further reducing actual usage costs.

Benchmarks published by Anthropic show that Fable 5.2 surpasses its own Opus 5.5 and the previous-generation Fable 5.1 in multiple tests. In the FrontierCode 2.0 programming test, Fable 5.2 scored 49.1% at the highest reasoning effort, significantly above Fable 5.1's 32.4% and also above Opus 5.5's 45.5%. In the FrontierSWE v2 test, the new model reached 69.9%, compared with 56.3% for Fable 5.1 and 62.3% for Opus 5.5.

In Humanity's Last Exam, which involves complex knowledge reasoning, Fable 5.2 with tool use scored 74.2%, higher than Fable 5.1's 65.6% and Opus 5.5's 67.7%. In the RoboDojo Sim dual-arm robot simulation test, the new model scored 64.9%, a marked improvement over the previous generation's 27.6%. In addition, in the Remote Labor Index test, Fable 5.2 scored 41.3%, while Fable 5.1 scored only 17.9%.

However, the new model does not lead in every test. For example, in the AutomationBench enterprise workflow test, Google Gemini 4 Carbon scored 53.9%, higher than Fable 5.2's 51.4%. These results also show that different models still differ in performance on specific task types, and overall capability cannot be judged from a single benchmark alone.

Anthropic also described Fable 5.2's actual performance in internal software development. The company says that more than 90% of merged code changes in its core codebase are now completed by Claude under human guidance. Engineering teams used Fable 5.2 to fix long-standing software defects that had been difficult to resolve, optimize the operating efficiency of multiple services, and complete some projects that had previously been difficult to launch due to cost or time constraints.

One case is particularly noteworthy: a group of AI agents using an early version of Fable 5.2 migrated a key production service from Python to Rust. After the migration, the service's startup speed increased by about 26 times, while operating efficiency and reliability also improved. Anthropic believes this reflects the new model's progress in engineering judgment and handling ambiguous requirements.

This release also involves adjustments to AI safety mechanisms. Fable 5.2 and Mythos 5.2 actually use the same underlying model, but different safety restrictions are configured for different user groups. The publicly available Fable 5.2 has safety protections in the cybersecurity and biology domains to block high-risk requests that could be misused. Anthropic says that compared with Fable 5.1, the new version had about 55% fewer safety blocks in early testing, meaning the model can handle normal requests more flexibly while still retaining restrictions on high-risk uses.

Mythos 5.2, by contrast, is open to vetted professional institutions, mainly in cybersecurity and life sciences research. Because its safety restrictions differ from the public version, ordinary users cannot access Mythos directly the way they can Fable. Anthropic hopes that through the vetting mechanism, it can allow professional researchers to carry out highly difficult work while reducing the risk of technology being misused.

On pricing, this upgrade also brings notable changes. The API prices for Fable 5.2 and Mythos 5.2 are both $8 per million input tokens and $40 per million output tokens, compared with $10 and $50 for the previous generation, a reduction of 20%. In addition, Anthropic offers an optional fast mode that can increase token generation speed by up to 2.5 times the normal mode, but the corresponding price rises to $16 per million input tokens and $80 per million output tokens.

Fable 5.2 is now available to Claude paid users and is offered through the API, with support for platforms including AWS, Google Cloud, and Microsoft Azure. Anthropic also announced that eligible paid subscribers will receive extra usage allowance this week, specifically one to three usage allowance resets depending on subscription length. Users on the Pro plan and standard seats of the Team plan can also apply up to 50% of their weekly usage allowance to Fable models.

Overall, the Fable 5.2 upgrade is not just a routine model performance iteration. Faster generation, lower invocation costs, and improved performance in complex programming and multi-step agent tasks mean Anthropic is further strengthening Claude's competitiveness in professional technical work. Mythos 5.2's continued restricted-release strategy also reflects the company's product approach of combining its strongest capabilities with safety review.

It should be noted that the above benchmark results and internal application cases all come from information published by Anthropic and cannot yet be fully equated with independent third-party test results in all actual usage scenarios.