Anthropic released Claude Fable 5.2 and Claude Mythos 5.2 on Monday, the newest versions of the frontier models it reserves for the most demanding technical work. The company says the upgraded pair improves on coding, reasoning, and vision, generates outputs up to 30% faster than Fable 5.1 and Mythos 5.1, and arrives at a lower price: $8 per million input tokens and $40 per million output tokens, with an optional fast mode that accelerates generation by up to 2.5 times for $16 per million input and $80 per million output tokens.

The launch quickly drew attention from the model-tracking community. Artificial Analysis, an independent AI evaluation site, scored Fable 5.2 at 65 on its intelligence index, the highest score it has recorded for any released model so far, and the announcement drew hundreds of comments on Hacker News within hours. For readers following the latest AI developments, this release marks the sharpest capability jump Anthropic has shown since the Fable line debuted in June.

One Model, Two Doors

Fable and Mythos are the same underlying model. The difference is access policy: Fable carries strong cybersecurity and biology safeguards that let Anthropic offer it to all users, while Mythos strips away fewer capabilities and is therefore restricted to organizations accepted into Anthropic's Life Sciences and Cyber Verification Programs. Anthropic said Fable 5.2's safeguards are as robust as Fable 5.1's but produce fewer unnecessary blocks, and that the new model includes the company's strongest-ever protections against distillation attacks.

The company also disclosed that prerelease safety testing included novel evaluations targeting specific concerning behaviors such as sandbox escapes, and that the model performed as well as or better than recent Anthropic models on most alignment measures. Full details are published in the Fable 5.2 and Mythos 5.2 system card.

What the Benchmarks Show

Anthropic positioned the new models as a step above Claude Opus 5.5, its recommended daily-driver model, on nearly every capability it evaluated. On FrontierCode 2.0, an agentic coding benchmark that measures whether a model's changes would actually be merged into open-source repositories, Fable 5.2 scored 49.1% at maximum reasoning effort, the highest score of any released model the company has tested. At high effort it reached 45.3%, roughly matching Opus 5.5's best score of 45.5% at about half the cost per task and beating GPT-6 Astra's best of 42.8%.

Other results from Anthropic's announcement:

  • FrontierSWE v2 (agentic coding): 69.9%, up from 56.3% for Fable 5.1 and ahead of Opus 5.5 (62.3%), GPT-6 Astra (65.5%), and Gemini 4 Carbon (61.2%).
  • Humanity's Last Exam with tools (multidisciplinary reasoning): 74.2%, versus 65.6% for Fable 5.1 and 57.2% for GPT-6 Astra at max effort.
  • Remote Labor Index (knowledge work): 41.3% at high effort, more than double Fable 5.1's 17.9% and well ahead of Opus 5.5's 21.3%.
  • OSWorld 2.1 (computer use): 89.3%, ahead of Opus 5.5 at 87.2%.
  • RoboDojo Sim (robotics): 64.9%, a jump from 27.6% for the previous version.
  • BenchCAD 2.1 (hardware engineering): 83.5 at max effort, versus 65.3 for Opus 5.5.

Gemini 4 Carbon held the edge in one area, scoring 53.9% to Fable 5.2's 51.4% on AutomationBench, a business-workflow benchmark. Mythos 5.2 posted the stronger cybersecurity and protein-design results, including 76.3% on CyScenarioBench and 89.6% on a CVE rediscovery test.

What the Model Did Before Launch

The most eye-catching claims in the announcement concern prerelease testing. Anthropic said Fable 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, and formalized the proof of Fermat's Last Theorem in 30 hours.

In one internal experiment, an Anthropic engineer asked the model to build a prototype operating system starting from an empty folder. The model coordinated more than 1,000 subagents over eight days and produced a system that boots on a laptop with its own kernel, disk, keyboard, and network drivers, permanent file storage, a windowing system, a login screen, and an installer. Anthropic also said a team of agents running an earlier version of the model migrated a critical production service from Python to Rust, and that the service now starts about 26 times faster.

The company added that more than 90% of changes merged into its own core codebase are now written by Claude with human guidance.

How Anthropic Suggests Using It

Despite the benchmark leap, Anthropic is not positioning Fable 5.2 as a replacement for its main lineup. For most coding and knowledge tasks it recommends starting with Opus 5.5 and moving up to Fable or Mythos only when a problem demands it, or down to Sonnet 5.5 and Haiku 5.5 when it doesn't. Fable 5.2 can also operate as an advisor that smaller models consult on hard problems, or as an orchestrator that delegates simpler tasks downward.

Early customers cited in the announcement include Vercel, Shopify, Datadog, Unity, Uber, Oracle, Snowflake, Cognition, and iGent. Joe Haddad, a member of technical staff at Vercel, said projects that had been pushed along with other models for weeks were finished and approved in days, while Tom Pritchard, a staff engineer at Shopify, described the model as feeling like working with another staff-level engineer.

Availability

Fable 5.2 is available now across Amazon Web Services, Google Cloud, and Microsoft Azure, and developers can call it as claude-fable-5-2 on the Claude Platform. Mythos 5.2 access runs through the verification programs, and Anthropic is also surfacing the model inside Claude Security for enterprise customers. For launch week, Anthropic is giving eligible paid subscribers extra Fable usage, including Fable access worth up to 50% of weekly usage limits on Pro plans.

The pricing cut is notable given the capability jump. Whether Fable 5.2's scores hold up under independent evaluation, and how quickly rivals OpenAI and Google respond with their next moves, will shape the frontier race heading into 2027.

---

Stay Ahead of AI

Get the latest AI news, analysis, and breakthroughs — all in one place.

Read more AI news →