Anthropic launches Claude Fable 5.2 and Mythos 5.2, up to 30% faster
Key Facts
Anthropic announced the launch of Claude Fable 5.2 and Claude Mythos 5.2 on November 9, 2026, saying both models have improved in coding, reasoning and vision. The company said they produce outputs up to 30% faster than their 5.1 versions, at lower prices for base usage. According to its announcement, pricing is $8 per million input tokens and $40 per million output tokens. The release includes two models that differ in their safeguards and access arrangements, although both are based on the same underlying model. Anthropic said Fable is available to all users, while access to Mythos is limited to specialized verification programs.
Anthropic says Fable includes strong cybersecurity and biology safeguards to protect against misuse, enabling it to be released to everyone. Mythos, by contrast, is available through trusted-access programs that require specialized verification. The company said Fable 5.2's safeguards in cybersecurity and biology are comparable to those of Fable 5.1, but result in fewer unnecessary refusals. It added that it has applied its strongest protections to date against distillation attacks. The company described safety controls and access management as central to the difference between the two versions.
On pricing, Anthropic said the base price for both models is $8 per million input tokens and $40 per million output tokens, compared with $10 and $50, respectively, for Fable 5.1. The company also offers an optional fast mode that accelerates token generation by up to 2.5 times. That mode costs $16 per million input tokens and $80 per million output tokens. For standard speed, the company says the models are up to 30% faster than their 5.1 versions. The new pricing pairs faster performance with lower base-use costs, with separate pricing for users who opt for the faster mode.
In agentic coding results published alongside the launch, Fable 5.2 scored 49.1% on the FrontierCode 2.0 test at the highest testing effort, versus 32.4% for Fable 5.1 and 45.5% for Opus 5.5. Anthropic said Fable 5.2 scored 45.3% at High effort, close to Opus 5.5's best result of 45.5% at Max effort. It added that task costs in the comparison at High effort were about half those of Opus 5.5. On the RoboDojo Sim test at Max effort, Fable 5.2 scored 64.9%, compared with 27.6% for Fable 5.1 and 29.8% for GPT-6 Astra. These results reflect tests published by Anthropic; the announcement did not include data on the market's reaction to the launch.
In a cybersecurity evaluation, Anthropic said its October run on 1,829 published vulnerabilities showed Mythos 5.2 outperforming Mythos 5.1 and Opus 5.5 at every effort level. The company cited a security firm as saying that one in six confirmed bugs found by Mythos 5.2 had not been found by the other models it tested. In a separate experiment, Anthropic said a team of 1,000 agents running Fable 5.2 completed the formal documentation of a proof of Fermat's Last Theorem in 30 hours. The company compared that with an earlier research attempt that took 11 days using an unreleased research version. These examples and comparison results are part of the company's evaluations published with the release.
The announcement follows a September 12, 2026 call by Anthropic Chief Executive Dario Amodei for the AI industry to slow development, giving safety measures time to catch up. Amodei warned then that AI could become capable within 6 to 12 months of directing a group of agents to take control of the internet. He said he believed slowing development by another year or two and using the time to improve alignment could greatly reduce the risk of a serious malfunction. The company also said it plans to provide independent evaluators with ongoing access to monitor its safety practices, and that it has begun implementing part of the plan. Separately, the AP reported that Anthropic and OpenAI were preparing for the possibility of entering the stock market at valuations that could reach hundreds of billions of dollars.