Anthropic's new Claude Fable 5.2 beat OpenAI's GPT-6 Astra, Google's Gemini 4 Carbon and Anthropic's own Opus 5.5 on almost every test the company ran. On the Remote Labor Index it nearly doubled every rival's score, and on a robot-control test it more than doubled GPT-6 Astra.
Fable 5.2 launched on Nov 9 alongside Claude Mythos 5.2. They are the same underlying model. Fable ships with cybersecurity and biology safeguards so anyone can use it; Mythos goes only to vetted organizations through Anthropic's Cyber and Life Sciences Verification Programs. Both generate output up to 30% faster than the 5.1 versions and cost less: $8 per million input tokens and $40 per million output tokens.
The scorecard
Here is Fable 5.2 against the next-best model on each of Anthropic's headline tests. All numbers are Anthropic's own, with safeguards on and at each model's highest available effort unless noted.
Test | Fable 5.2 | Next best |
|---|---|---|
Agentic coding (FrontierSWE v2) | 69.9% | 65.5% GPT-6 Astra |
Code that gets merged (FrontierCode 2.0) | 49.1% | 45.5% Opus 5.5 |
Scientific research (Terminal-Bench-Science) | 74.6% | 68.1% GPT-6 Astra |
Knowledge work (Remote Labor Index) | 41.3% | 21.3% Opus 5.5 |
Robot control (RoboDojo Sim) | 64.9% | 39.4% Opus 5.5 |
Hardware design (BenchCAD 2.1, 0-100) | 83.5 | 65.3 Opus 5.5 |
Expert reasoning (Humanity's Last Exam, with tools) | 74.2% | 67.7% Opus 5.5 |
Agentic search (WANDR) | 82.1% | 72.3% Opus 5.5 |
Computer use (OSWorld 2.1) | 89.3% | 87.2% Opus 5.5 |
Chart reading (Chartography) | 73.4% | 72.5% Gemini 4 Carbon |
Business workflows (AutomationBench) | 51.4% | 53.9% Gemini 4 Carbon |
It isn't a clean sweep. Gemini 4 Carbon won AutomationBench, and Opus 5.5 edged Fable 5.2 on the Elo rating of the AA-Briefcase knowledge-work test, 1800 to 1785.
Coding: 1,000 agents, eight days, one operating system
On FrontierSWE v2, Fable 5.2 scored 69.9%, ahead of GPT-6 Astra at 65.5% and well clear of its predecessor, Fable 5.1, at 56.3%.

Anthropic says Claude now writes more than 90% of the changes merged into its core codebase, with human guidance. A team of agents running an early version of Fable 5.2 moved a critical production service from Python to Rust, and the service now starts about 26 times faster.
The boldest test: one engineer asked Fable 5.2 to build a prototype operating system starting from an empty folder. The model coordinated more than 1,000 subagents around the clock for eight days and produced a system that boots on a laptop, complete with drivers, file storage, windows, a login screen and an installer.
Early testers describe the same appetite for long, unattended work. Uber says Fable 5.2 ran a 10 to 12 hour security audit, sent out about 170 reviewer agents and came back with fix pull requests. Oracle says it wrote correct SQL for a debt-analysis query where two other models returned totals 5 to 8 times too high.
Real work and real robots
The biggest jumps came outside pure coding. On the Remote Labor Index, Fable 5.2 scored 41.3% at High effort. Opus 5.5 managed 21.3% and GPT-6 Astra 20.8%.

Robotics shows an even sharper break. On RoboDojo Sim, which tests control of a simulated two-arm robot from camera images, Fable 5.2 scored 64.9%, more than double GPT-6 Astra's 29.8% and Fable 5.1's 27.6%. Anthropic says it also beat most AI systems built specifically to control robots, and its demo shows Fable-driven robots solving a puzzle cube and unscrewing a bottle cap.

Hardware followed. From a one-paragraph prompt, Fable 5.2 designed three working devices, a voice-assistant hub, a USB speakerphone and a handheld virtual pet, including their circuit boards, 3D-printed cases and software. Each board was ready to manufacture in under a day, and all three worked without electrical fixes.
Science: black holes and Fermat
Anthropic says Fable 5.2 solved two open problems in theoretical physics, both about black holes. One was a calculation laid out in 2003 but too laborious for anyone to finish; the model completed it largely on its own. Both results are out as preprints.
Math offers the cleanest proof point. Anthropic's first computer-checked formalization of Fermat's Last Theorem took an unreleased research model 11 days, a custom harness and a human guide. About 1,000 Fable 5.2 agents did it in 30 hours with minimal human help, using a third of the action tokens and producing a proof seven times shorter. Opus 5.5, in the same setup, got 26% of the way in 18 hours and then stalled.

In biology, Mythos 5.2 found a new way to predict the full protein shell of the viruses used to deliver gene therapies, accurately modeling 28 of the 34 shells it attempted.
Mythos: the security specialist
Mythos 5.2 is Anthropic's strongest cybersecurity model. On CVE Rediscovery, an internal test built on 1,829 real, published vulnerabilities, it found 89.6%. That is 10.2% more than Opus 5.5 and 14.4% more than Mythos 5.1.

Mozilla, which has used Claude to harden Firefox all year, says Mythos 5.2 roughly doubled Mythos 5.1's vulnerability-discovery numbers. One security company found that one in six confirmed bugs caught by Mythos 5.2 were missed by every other frontier model it tested. Fable 5.2, by design, blocks most of this work.
Safety: fewer escapes, fewer false alarms
In an adversarial test of more than 1,800 scenarios with a deliberately weak sandbox, Mythos 5.2 used credentials or files planted outside it in 3.1% of cases. Mythos 5 did so in 39% and Mythos 5.1 in 21%.

Fable's safety filters also got less jumpy, with about 55% fewer unnecessary blocks than Fable 5.1 in early testing. Anthropic says its anti-distillation defenses cut the data harvested by illicit copiers by 90% to 99%.
Anthropic flags one caveat itself: both models voice suspicion that they are being tested more often than any recent Claude model, which makes their real-world behavior harder to predict. External testers included the UK AI Security Institute and METR.
Price, access and the catch
Fable 5.2 is available now on all platforms, including AWS, Google Cloud and Microsoft Azure. An optional fast mode runs up to 2.5 times faster for $16 per million input tokens and $80 per million output tokens. Mythos 5.2 requires approval through Anthropic's verification programs.
The catch sits in Anthropic's own advice: it still tells most users to start with Opus 5.5 and step up to Fable only for the most ambitious, open-ended work. Every number here also comes from Anthropic, and several tests are internal, so independent runs will show how much of the gap holds.
The headline isn't any single score. It's that Anthropic's top model got better, faster and cheaper in the same release.
Get stories like this in your inbox every week. Subscribe to The Week in Perspective.
Sources: Anthropic announcement

