Anthropic announced on November 9, 2026 the launch of Claude Fable 5.2 and Claude Mythos 5.2, a matched pair of upgrades built, in the company's words, to tackle "the most daunting technical challenges." The two names share a single underlying model: Fable carries cybersecurity and biology safeguards that let Anthropic offer it to all users, while Mythos is reserved for vetted organizations and is available only through Anthropic's recently expanded Life Sciences and Cyber Verification Programs.
According to Anthropic's announcement, the new models improve on their 5.1 predecessors in coding, reasoning, and vision, and generate output up to 30% faster. They also cost less: Fable 5.2 is priced at $8 per million input tokens and $40 per million output tokens. An optional fast mode accelerates token generation by up to 2.5 times at $16 per million input tokens and $80 per million output tokens. For more context on this story, see our ongoing breaking AI news.
One model, two doors
The dual-identity arrangement is unusual among frontier AI labs. Anthropic says Fable 5.2's cybersecurity and biology safeguards exist "to guard against misuse," which is what allows unrestricted availability. Mythos 5.2, by contrast, is gated: organizations pursuing beneficial applications in cybersecurity or biomedical research can apply for access through the verification programs.
Testers cited by Anthropic also report that both versions communicate more clearly than the previous Fable and Mythos models, a recurring complaint about highly capable models that produce dense or evasive explanations.
What the models did before launch
The most striking claims in the announcement come from pre-release testing. Anthropic says the model behind Fable 5.2 and Mythos 5.2 helped its engineers complete otherwise-infeasible software migrations, designed three working electronic devices from scratch, solved two open problems in the physics of black holes, made substantial progress on robotics evaluations, and formalized the proof of Fermat's Last Theorem in 30 hours.
Those claims are Anthropic's own and have not been independently verified, but they signal how the company is positioning the release: as a research and engineering instrument rather than a conversational upgrade.
Benchmark results, according to Anthropic
Anthropic's published benchmark table places Fable 5.2 above the company's own Claude Opus 5.5 in nearly every capability it evaluated. On FrontierCode 2.0, an agentic coding benchmark that measures whether a model's code changes would be merged by open-source maintainers, Fable 5.2 scored 49.1% at maximum effort — which Anthropic says is the highest score of any released model it has tested. Claude Opus 5.5 scored 45.5%, while the comparison figures listed for OpenAI's GPT-6 Astra and Google's Gemini 4 Carbon were 42.8% and 41.1% respectively.
Other results in Anthropic's table include:
- FrontierSWE v2 (agentic coding): 69.9% for Fable 5.2, up from 56.3% for Fable 5.1
- Humanity's Last Exam with tools (multidisciplinary reasoning): 74.2%, versus 65.6% for the previous generation
- OSWorld 2.1 (computer use): 89.3%, ahead of Opus 5.5 at 87.2%
- RoboDojo Sim (robotics): 64.9%, a sharp jump from Fable 5.1's 27.6%
- Remote Labor Index (knowledge work): 41.3% at high effort, compared with 17.9% for Fable 5.1
The gated Mythos 5.2 configuration posted the stronger safety-relevant numbers: 76.3% on CyScenarioBench, 89.6% on a CVE rediscovery test, and 85.1% on de novo protein binder design over a 24-hour run, in each case with safeguards enabled.
Anthropic's own cost-efficiency analysis may matter as much as the raw scores. The company reports that Fable 5.2 at high effort scored roughly level with Opus 5.5's best result on FrontierCode 2.0 while costing about half as much per task, and that on medium effort — the default in Claude apps and Claude Code — it completes challenging tasks with fewer tokens than Fable 5.1.
Safety testing and safeguards
Anthropic says it ran extensive pre-release safety testing on the shared model, including novel evaluations designed to probe specific concerning behaviors such as sandbox escapes. The company states that high-stakes risks from misalignment remain low and that the model performed as well as, or better than, its recent models on most alignment measures.
Fable 5.2's cyber and biology safeguards are described as robust as those of Fable 5.1 but with fewer unnecessary blocks, and Anthropic says it implemented its strongest-ever protections against distillation attacks — attempts by rival developers to train competing models on a target model's outputs. Full details are published in the Claude Fable 5.2 and Mythos 5.2 System Card.
Where Fable 5.2 fits in the Claude lineup
Anthropic is explicit that Fable and Mythos are not meant to replace the everyday models. For most coding and knowledge tasks, the company recommends starting with Claude Opus 5.5, which it calls its "strongest daily driver," then moving up to Fable or Mythos for unusually hard work, or down to Claude Sonnet 5.5 or Claude Haiku 5.5 for routine requests.
Fable 5.2 can also operate as an Advisor — consulted by Haiku, Sonnet, or Opus models when they hit harder problems — or as an Orchestrator that delegates simpler tasks to smaller models. That positions the release as much about multi-model workflows as single-model performance.
What it means for the AI race
If the reported numbers hold up under external scrutiny, Fable 5.2 narrows the gap between Anthropic's specialist models and the flagship offerings of OpenAI and Google DeepMind at a lower price point. The timing is notable for another reason too: the launch lands one day before Anthropic launched the roadshow for its initial public offering, giving the company a capability showcase to point to as investors weigh the most anticipated technology listing in years.
For users, availability is immediate — Fable 5.2 for everyone, and Mythos 5.2 for organizations that clear the verification bar.
---
Stay Ahead of AIGet the latest AI news, analysis, and breakthroughs — all in one place.
Read more AI news →
