This article may contain affiliate links. Learn more on our Partner Program page.
Just over two months after Fable 5.1, Anthropic is releasing the next version of its most powerful generally available model. Fable 5.2 is said to be faster, outperform Opus 5.5 in almost all tests, and for the first time comes with lower base prices for input and output.
Anthropic unveiled Claude Fable 5.2 and Claude Mythos 5.2 on November 9th. As with their predecessors, they use the same model but with different protection mechanisms: Fable is open to all users and blocks sensitive requests related to cybersecurity and biology, while Mythos is only available to vetted organizations in these fields.
The series began in June with Fable 5, the first Mythos model for the general public, followed in September by Fable 5.1 with lower costs for cached content. With version 5.2, the base price is now also reduced.
Key Facts at a Glance
- Fable 5.2 and Mythos 5.2 have been available since November 9th; Fable on all platforms, Mythos via testing programs and for enterprise customers in Claude Security.
- Input costs are $8 instead of $10 per million tokens, output costs are $40 instead of $50.
- According to Anthropic, responses come up to 30 percent faster than in version 5.1, and an optional fast mode accelerates the process by up to 2.5 times.
- The security filters block approximately 55 percent less often than in Fable 5.1.
- For most tasks, Anthropic still recommends Opus 5.5 as a starting point.
Where Fable 5.2 improves
Anthropic cites programming, logical reasoning, and visual comprehension as key areas of focus. In the company's comparison chart, Fable 5.2 outperforms Opus 5.5 in almost all disciplines, in some cases significantly.
| Test procedure | Fable 5.2 | Fable 5.1 | Opus 5.5 |
|---|---|---|---|
| Agent Programming (FrontierCode 2.0) | 49,1 % | 32,4 % | 45,5 % |
| Agent Programming (FrontierSWE v2) | 69,9 % | 56,3 % | 62,3 % |
| Business processes (AutomationBench) | 51,4 % | 31,4 % | 42,5 % |
| Computer operation (OSWorld 2.1) | 89,3 % | 83,5 % | 87,2 % |
| Interdisciplinary reasoning (Humanity's Last Exam, with tools) | 74,2 % | 65,6 % | 67,7 % |
| Robot control (RoboDojo Sim) | 64,9 % | 27,6 % | 39,4 % |
The model isn't at the top in every category. In the same table, Google's Gemini 4 Carbon achieves 53.9 percent in business processes. However, the robot control aspect is noteworthy, where the score more than doubles compared to Fable 5.1.
What Anthropic has built with it
The announcement is supported by a number of real-world examples. A team of agents with a pre-release version migrated a production service from Python to Rust, which then started approximately 26 times faster. In an experiment, Fable 5.2 coordinated over 1,000 sub-agents and wrote an operating system that boots on a laptop in eight days.
In addition, there are results beyond programming: The model has solved two open problems in theoretical physics concerning black holes and, using around 1,000 parallel agents, formalized the proof of Fermat's Last Theorem in 30 hours. Opus 5.5, with the same setup, achieved about a quarter of that in 18 hours.
Among the testers, Cognition commented on a weakness that many users of language models are familiar with:
Claude Fable 5.2 says that it doesn't know something, rather than hallucinating.
Myth 5.2 for security research
Anthropic describes Mythos 5.2 as its strongest cybersecurity model. In an internal test of 1,829 known vulnerabilities, it found 10.2 percent more than Opus 5.5 and 14.4 percent more than Mythos 5.1. Access is available through the Cyber Verification Program, and for the life sciences, through a dedicated testing program. Enterprise customers also use the model in Claude Security, a service that scans code for vulnerabilities.
Mozilla states that it has been working with Claude models all year to secure Firefox:
Mythos 5.2 has roughly doubled the key performance indicators (KPIs) for detecting vulnerabilities compared to Mythos 5.1 in the respective test runs.
Fable 5.2 itself still blocks most security tasks. If a filter is triggered, Opus 4.8 responds for cybersecurity and Opus 5 for biology. However, initial tests in claude.ai, Claude Code, and Cowork show that the filters are triggered approximately 55 percent less frequently than in Fable 5.1, about 60 percent less frequently for cybersecurity, and 40 percent less frequently for biology.
Prices and availability
Fable 5.2 is now available on all platforms, including Amazon Web Services, Google Cloud, and Microsoft Azure. The model is named claude-fable-5-2 via the API.
| Price per million tokens | Fable 5.2 | Fable 5.1 |
|---|---|---|
| Input | 8 US dollars | 10 US dollars |
| Output | 40 US dollars | 50 US dollars |
| Quick input | 16 US dollars | – |
| Output in fast mode | 80 US dollars | – |
Both base prices will thus decrease by 20 percent. To celebrate the launch, there's a special offer: subscribers on eligible paid plans will receive additional access this week, with one to three Fable limits reset depending on the subscription length. In the same announcement, Anthropic states that for the Pro plan and Standard slots in the Team plan, up to 50 percent of the weekly Fable usage limit will be reset.
Why price is more important than the table
With Fable 5.1, the list prices remained unchanged; only the replay of cached content became cheaper. With version 5.2, both base prices have dropped for the first time, and this for a model that performs better in most tests. I think that's the real news: those who use Fable via the API pay less for more performance.
However, Anthropic's own recommendation remains telling. For most programming and knowledge-based tasks, users should continue to start with Opus 5.5 and only switch to Fable for particularly demanding technical projects. Those who primarily use Claude for writing, research, or small scripts will therefore likely notice little difference from the new top-of-the-line model.
Do you already use Fable for your own projects in everyday life, or is Opus sufficient for almost everything you do with Claude? Let us know in the comments which tasks have made switching to Fable worthwhile for you so far.
