Anthropic launches Claude Fable 5.2 and Mythos 5.2, improving coding

Claude Fable 5.2 and Claude Mythos 5.2 have been released by Anthropic, offering faster, cheaper performance on coding, vision and scientific reasoning tasks while adding stronger safety safeguards. The upgrade matters because the models claim to handle projects that were previously infeasible, from large‑scale code migrations to formal proofs, and they are now accessible through Anthropic’s expanded Life Sciences and Cyber Verification programs.

What are Claude Fable 5.2 and Claude Mythos 5.2?

Anthropic describes the two models as "the same underlying model" with different usage constraints. Fable includes built‑in cybersecurity and biology safeguards that let Anthropic make it broadly available. Mythos carries the same core capabilities but is gated behind the Life Sciences and Cyber Verification programs for organizations that need tighter controls on biomedical or security‑sensitive work. Both versions are positioned for "the most ambitious, open‑ended technical work" according to the company’s announcement.

How do the new models perform compared to previous versions?

During prerelease testing, Anthropic measured the models on a suite of benchmarks. Across the board, Fable 5.2 and Mythos 5.2 outperformed their 5.1 predecessors and the daily‑driver model Claude Opus 5.5. Highlights include:

• Agentic coding (FrontierCode 2.0) – Fable 5.2 reached 49.1 % accuracy at Xhigh effort, the highest score of any released model, and matched Opus 5.5 at High effort for roughly half the cost.
• Hardware engineering (BenchCAD 2.1) – At Max effort, Fable 5.2 scored 83.5, well above Opus 5.5’s 65.3.
• Scientific research (Terminal‑Bench‑Science) – Mythos 5.2 posted a 74.6 % score, beating Opus 5.5’s 75.3 % only when safeguards were disabled, indicating strong reasoning even with safety layers.
• Cybersecurity (CyScenarioBench) – Mythos 5.2 achieved 76.3 % versus 61.7 % for Mythos 5.1, showing a sizable gain in threat‑modeling tasks.
• Visual chart recognition (Chartography) – Fable 5.2 recorded 73.4 % accuracy, surpassing Opus 5.5’s 64.4 %.

The company also notes that token‑generation speed is up to 30 % faster than the 5.1 series, and an optional fast mode can accelerate output by up to 2.5×.

Which use cases does Anthropic highlight for the models?

Anthropic cites a range of high‑impact applications that were completed during internal testing:

• Software migrations – A team used an earlier Fable 5.2 build to rewrite a critical production service from Python to Rust, achieving a 26‑fold speed increase.
• Operating‑system prototype – The model coordinated more than 1,000 sub‑agents over eight days to produce a bootable OS with drivers, a login screen and built‑in Claude functionality. The demo video is available here: Operating‑system prototype video.
• Electronic‑device design – Engineers designed three working electronic devices from scratch using the model’s reasoning abilities.
• Physics research – The models helped solve two open problems in black‑hole physics and contributed to a 30‑hour formal proof of Fermat’s Last Theorem.
• Robotics evaluation – Substantial progress was reported on simulated robotics tasks (RoboDojo Sim).
• Security audits – A 10‑12‑hour audit generated 170 reviewer agents, pull‑request fixes and a policy proposal, with the model continuing until the job was truly complete.

User testimonials from companies such as Vercel, Shopify, Datadog, Unity and Uber echo the internal results, emphasizing the model’s ability to understand intent without repeated clarification, to work on ambiguous systems, and to iterate on complex game‑dev or security tasks.

What are the pricing and speed options?

Anthropic lists the following rates for the publicly available Fable 5.2 and Mythos 5.2:

• Standard mode – $8 per million input tokens and $40 per million output tokens.
• Fast mode – $16 per million input tokens and $80 per million output tokens, delivering up to 2.5 × faster generation.

Both pricing tiers are lower than the previous 5.1 models, and the speed advantage is reinforced by the 30 % token‑generation improvement noted above.

How can Grafy users experiment with these models?

Grafy’s Code studio lets developers run AI‑assisted coding sessions with any model available through OpenRouter, Vertex AI, Replicate and other providers. Because Anthropic’s Claude models are part of the OpenRouter catalog, you can spin up a Claude Fable 5.2 or Mythos 5.2 instance directly inside Grafy Code and test the same agentic workflows described by Anthropic.

1. Open Grafy Code – [/code] – and choose Claude Fable 5.2 from the model selector.
2. Paste a coding brief (e.g., “migrate this Python module to Rust”) and let the model generate a pull request.
3. Use the built‑in terminal to run the generated code, inspect token usage, and compare cost against the $8/$40 pricing.
4. For safety‑sensitive projects, switch to Mythos 5.2 after applying for access via Anthropic’s Life Sciences or Cyber Verification programs.

If you prefer a conversational interface, Grafy Chat also supports Claude models, allowing you to explore vision or scientific‑reasoning prompts without writing code first. Finally, the detailed System Card released by Anthropic can be loaded into Grafy Reader for searchable reference, helping you stay informed about the model’s safeguards and evaluation methodology.

To explore this yourself, open Grafy Code: an AI coding studio.

Frequently asked questions

How does Claude Fable 5.2’s coding ability compare to Claude Opus 5.5?

On the FrontierCode 2.0 benchmark, Fable 5.2 matches Opus 5.5’s top score at High effort while costing roughly half as much per task. At Xhigh effort, Fable 5.2 actually exceeds Opus 5.5, achieving 49.1 % accuracy, the highest among tested models.

What safety measures are new in Fable 5.2 and Mythos 5.2?

Anthropic added novel sandbox‑escape evaluations and the strongest‑ever protections against distillation attacks. The cyber and biology safeguards remain as robust as in 5.1 but generate fewer unnecessary blocks, according to the company’s safety testing results.

Can I use Claude Mythos 5.2 for biomedical research without a special program?

Mythos 5.2 is only available to organizations that have been approved through Anthropic’s Life Sciences or Cyber Verification programs. Researchers must apply for access; otherwise, the publicly available Fable 5.2, which includes similar core capabilities but broader safeguards, should be used.

---

Source: Anthropic announcement, November 9 2026. For the full system card and additional benchmark details, see the Claude Fable 5.2 and Mythos 5.2 System Card linked in the original release.

---

Source: Hacker News — front page — Claude Fable 5.2 and Mythos 5.2