Anthropic upgrades Claude Fable and Mythos to version 5.2 with faster speeds and lower prices

What's new

Claude Fable 5.2 and Claude Mythos 5.2 are upgrades to Anthropic's top-tier technical models, both now generating output up to 30 percent faster than their previous version, Claude Fable 5.1 and Claude Mythos 5.1 ¹. The pricing also dropped: standard mode costs $8 per million input tokens and $40 per million output tokens, which means a task that once cost $100 could now cost around $70 . An optional "fast mode" is available for double that price, speeding token generation by up to 2.5 times .

Fable and Mythos share the same base model, but Fable carries stricter safeguards against misuse in areas like cybersecurity and biology, making it safe for general use. Mythos removes some of those restrictions and is available through Anthropic's Life Sciences and Cyber Verification Programs for approved organizations pursuing beneficial applications ¹². Earlier Wikipedia reporting noted that industry estimates put Claude Mythos at roughly 8 trillion parameters and Claude Fable 5 at around 5 trillion ³.

In prerelease testing, these tools enabled our engineers to carry out software migrations that would otherwise have been impractical, created three fully functional electronic gadgets from the ground up, addressed two longstanding questions in black hole physics, made significant advances in robot performance assessments, and produced a verified mathematical proof of Fermat's Last Theorem within 30 hours—as well as supporting numerous other uses detailed later ¹.

How well do they perform?

On benchmark tests, Fable 5.2 scores above Claude Opus 5.5 in almost every category ¹. Key results include:

• FrontierCode 2.0 (agentic coding): 49.1 percent at high effort
• FrontierSWE v2 (software engineering agent): 69.9 percent
• CyScenarioBench (cybersecurity): 76.3 percent for Mythos 5.2
• CVE Rediscovery: 89.6 percent for Mythos 5.2
• OSWorld 2.1 (computer use): 89.3 percent
• Humanity's Last Exam (multidisciplinary reasoning): 74.2 percent

At medium effort, Fable 5.2 uses fewer tokens than its predecessor, and at high effort it matches Opus 5.5 at maximum effort for roughly half the cost ¹.

Safety checks

Anthropic ran extensive safety testing before release, developing new evaluations focused on behaviors like sandbox escapes — where an AI tries to break out of its own controlled environment ¹. The company says high-stakes risks from misalignment remain low, and Fable 5.2's safeguards produce fewer unnecessary blocks while adding the strongest protections yet against distillation attacks, in which someone tries to copy a model's behavior by feeding it prompts .

This story reports on a trend for information only. It is not advice or encouragement to do any of it.

Key facts

• Claude Fable 5.2 and Mythos 5.2 generate outputs up to 30% faster than version 5.1
• Standard pricing: $8 per million input tokens, $40 per million output tokens; fast mode costs double
• Fable 5.2 scores above Claude Opus 5.5 on most benchmarks tested
• Mythos 5.2 requires approval through Anthropic's Life Sciences and Cyber Verification Programs

Sources

1. Introducing Claude Fable 5.2 and Claude Mythos 5.2
2. Life sciences verification program
3. Claude Mythos — en.wikipedia.org