Anthropic released Claude Fable 5.2 and Claude Mythos 5.2 on Monday, November 9, 2026, and the news was reported widely on Tuesday, November 10. They are the company’s newest models for difficult technical work, and they arrive as faster and cheaper successors to the 5.1 versions. Both models share the same underlying model, but they are aimed at different users and carry different safeguards. Fable 5.2 is broadly available to everyone. Mythos 5.2 has fewer safety blocks and is restricted to vetted cybersecurity and life sciences organizations through Anthropic verification programs. Pricing for both is 8 dollars per million input tokens and 40 dollars per million output tokens, down from 10 and 50 dollars, so about 20 percent cheaper. Outputs are up to 30 percent faster than the 5.1 versions.
Quick Facts
| Detail | What to know |
|---|---|
| Launch date | Monday, November 9, 2026 |
| Models | Claude Fable 5.2 and Claude Mythos 5.2 |
| Who can use Fable 5.2 | Broadly available to everyone |
| Who can use Mythos 5.2 | Vetted cybersecurity and life sciences organizations only |
| Input price | 8 dollars per million tokens, down from 10 dollars |
| Output price | 40 dollars per million tokens, down from 50 dollars |
| Speed | Outputs up to 30 percent faster than the 5.1 versions |
| Coding benchmark | Fable 5.2 scored 49.1 percent on FrontierCode 2.0 at Xhigh effort |
| Security benchmark | Mythos 5.2 identified 89.6 percent of vulnerabilities in internal testing |
| Developer model ID | claude-fable-5-2 |
| Where to find them | Anthropic platforms, Amazon Web Services, Google Cloud and Microsoft Azure |
What Anthropic Actually Released
The launch covers two models that sit on the same foundation. That shared base is an important detail, because it means the difference between Fable 5.2 and Mythos 5.2 is not raw ability. It is access and safeguards. Anthropic built one strong model for hard technical work and then shaped it into two products for two very different audiences.
Fable 5.2 is the general release. It is the model most developers, teams and businesses will use. It carries the standard safety systems that Anthropic applies to its public products, and it is available through the company’s own platforms and through the big cloud providers.
Mythos 5.2 is the specialist release. It has fewer safety blocks, and that change is deliberate. Anthropic says the looser limits let the model work more freely on defensive cybersecurity tasks and on life sciences research where normal guardrails can get in the way of legitimate work. Access is not open. Only organizations that pass through Anthropic verification programs can use it, and the focus is on vetted cybersecurity and life sciences groups. That split lets Anthropic serve advanced security teams without handing a less restricted model to the general public.
The launch comes in the same week Anthropic set a planned IPO price range of 70 to 78 dollars per share.
Faster Output and Lower Prices
Speed and price are the two changes that most users will notice first. Anthropic says outputs are up to 30 percent faster than the 5.1 versions. For people who run long coding sessions or large research jobs, that kind of speed gain means more iterations and faster debugging.
The price cut is just as direct. Both models now cost 8 dollars per million input tokens and 40 dollars per million output tokens. The earlier prices were 10 dollars for input and 50 dollars for output. That is about a 20 percent drop on both sides of the bill.
There is also a fast mode for users who need even more speed. Fast mode costs 16 dollars per million input tokens and 80 dollars per million output tokens. It is available on Claude Code, the Claude Platform, Google Cloud and Microsoft Foundry. Mythos 5.2 fast mode is limited to organizations in the verification programs.
How Fable 5.2 Performs on Coding
Benchmark numbers need care, because a single score never tells the full story. Still, the figures Anthropic published for Fable 5.2 are strong, and they help explain why the company is presenting this as a step forward rather than a small update.
On the FrontierCode 2.0 benchmark, Fable 5.2 scored 49.1 percent at its Xhigh effort setting. The earlier Fable 5.1 scored 32.4 percent on the same test. That is a large jump between two versions, and it suggests real gains on long, multi step coding problems.
Anthropic also says that at its High effort setting, Fable 5.2 reached a coding benchmark score comparable to Opus 5.5 at its highest setting, at roughly half the cost per task.
Fable 5.2 also took the No. 1 spot on the Artificial Analysis Intelligence Index. Sources disagree on the exact index score number, so this article does not state one.
In early testing across claude.ai, Claude Code and Cowork, Anthropic recorded about 55 percent fewer blocks with Fable 5.2 than with Fable 5.1. A block is when the model refuses or stops a request because of a safety rule.
What Mythos 5.2 Does Differently
Mythos 5.2 is built for a narrow and sensitive job. Anthropic positions it as a model for defensive cybersecurity work, where finding weaknesses before attackers do is the whole point. The headline number comes from Anthropic’s internal CVE Rediscovery evaluation, which tested the model against 1,829 previously published vulnerabilities.
In that test, Mythos 5.2 identified 89.6 percent of the vulnerabilities. Mythos 5.1 identified 78.3 percent, and Opus 5.5 with safeguards disabled identified 81.3 percent. The gap between Mythos 5.2 and the earlier version is wide, and it also sits ahead of Opus 5.5 even when that model’s safeguards were turned off. For security teams, that kind of recall matters. Missing a known weakness is worse than flagging one that turns out to be minor, because the missed flaw is the one an attacker can use.
These are Anthropic’s own internal test results, and they should be read that way. The company designed the evaluation, ran it on its own models, and published the scores. Independent testing may tell a different story over time.
The life sciences side of Mythos 5.2 follows the same logic. Some research tasks need a model that can discuss sensitive biology in depth, and standard guardrails can block legitimate scientific work.
Big Claims About Agent Teams
Beyond the benchmarks, Anthropic shared some striking claims about what large teams of AI agents can do with these models. According to the company, teams of more than 1,000 AI agents using the models built a working operating system from scratch in eight days, solved two previously open problems in black hole physics, and designed three working electronic devices.
These are Anthropic’s own claims, not independently verified facts. That distinction matters, because the claims are extraordinary and the details behind them are limited in public reporting. Until that evidence appears, readers should treat the claims as the company’s account of its internal tests, not as settled fact.
Where You Can Use the New Models
Availability is broad for Fable 5.2. It is available now on Anthropic platforms and through Amazon Web Services, Google Cloud and Microsoft Azure. That spread matters, because many companies already run their workloads on one of those clouds and prefer to use AI models inside the same environment. The developer model ID is claude-fable-5-2, which is the name developers will use when they call the model through an API.
Mythos 5.2 follows the same cloud path, but only for approved organizations. The restriction is not a technical limit. It is a policy choice, enforced through the verification programs. A cybersecurity firm or life sciences lab that wants Mythos 5.2 will need to go through that process first. For everyone else, Fable 5.2 is the available option, and it carries the full safety systems.
The launch also fits into a busy week for AI news. Readers who want wider context can look at the magazine’s coverage of Anthropic’s planned IPO price range, the new Google TPU chips for AI workloads and Australia’s new AI safety rules. Together those stories show an industry moving fast on models, hardware and regulation at the same time, and the Fable 5.2 and Mythos 5.2 launch sits in the middle of that wider shift.
FAQs
What is the difference between Claude Fable 5.2 and Mythos 5.2?
Both share the same underlying model, but Fable 5.2 is broadly available with standard safeguards, while Mythos 5.2 has fewer safety blocks and is restricted to vetted cybersecurity and life sciences organizations through Anthropic verification programs.
How much do Claude Fable 5.2 and Mythos 5.2 cost?
Both cost 8 dollars per million input tokens and 40 dollars per million output tokens, down from 10 and 50 dollars, which is about 20 percent cheaper than the 5.1 versions. Fast mode costs 16 dollars per million input tokens and 80 dollars per million output tokens.
How fast are the new models?
Anthropic says outputs are up to 30 percent faster than the 5.1 versions. Fast mode is available for users who need even more speed on Claude Code, the Claude Platform, Google Cloud and Microsoft Foundry.
Where can I use Claude Fable 5.2?
Fable 5.2 is available now on Anthropic platforms and through Amazon Web Services, Google Cloud and Microsoft Azure. The developer model ID is claude-fable-5-2.
SOURCES
Final Thoughts
Claude Fable 5.2 and Mythos 5.2 are a clear statement of intent from Anthropic. Faster output, lower prices and stronger benchmark scores make Fable 5.2 a serious option for coding and research teams, while Mythos 5.2 gives vetted security and life sciences groups a more open tool for sensitive work. The split between the two models shows how Anthropic is trying to balance power and control. One model for everyone, and one model for a small group that has been checked first.
The benchmark numbers are strong, and the price cut makes the models easier to adopt at scale. The big agent team claims are far bolder, and they need independent evidence before anyone treats them as fact. Developers can try Fable 5.2 today through Anthropic platforms or the major clouds, and the real test will be how it performs on their own projects over the coming weeks.
