Anthropic Unleashes Claude Fable 5.2 and Mythos 5.2, Crushing GPT-6 Astra and Gemini 4 Carbon

Anthropic has launched Claude Fable 5.2 and Claude Mythos 5.2, introducing improvements in coding, reasoning, scientific research, and other complex tasks.

The company says its new AI models generate responses up to 30% faster than their predecessors while offering 20% lower API prices. Anthropic also claims the models outperform competing systems from OpenAI and Google in several benchmarks.

Claude Fable 5.2 delivers faster performance at lower prices

Both models share the same architecture but offer different levels of access and safety restrictions.

Claude Fable 5.2 targets general users and includes stricter safeguards for cybersecurity and biological research. Claude Mythos 5.2 offers fewer restrictions to verified organizations conducting cybersecurity and biomedical research.

Anthropic has also reduced API pricing by 20%, bringing input costs down from $10 to $8 per million tokens and output costs from $50 to $40 per million tokens.

The models generate output up to 30% faster than Fable 5.1 and Mythos 5.1. Developers can also enable an optional fast mode that delivers up to 2.5 times faster token generation, although it doubles the standard API price.

Claude Fable 5.2 outperforms GPT-6 Astra and Gemini 4 Carbon in benchmarks

According to Anthropic’s benchmark results, Claude Fable 5.2 achieved 49.1% on FrontierCode 2.0, surpassing Claude Opus 5.5 at 45.5%, GPT-6 Astra at 42.8%, and Gemini 4 Carbon at 41.1%.

That also represents a substantial improvement over Fable 5.1, which scored 32.4% on the same benchmark.

On Humanity’s Last Exam with tools, Fable 5.2 reached 74.2%, compared with 57.2% for GPT-6 Astra.

Anthropic also reported strong results across several other tests, including:

• FrontierSWE v2: Fable 5.2 scored 69.9% in agentic coding tasks.
• OSWorld 2.1: The model achieved 89.3% in computer-use tasks.
• Terminal-Bench-Science: Fable 5.2 scored 74.6%, while Mythos 5.2 reached 75.3%.
• CVE Rediscovery: Mythos 5.2 achieved 89.6% in vulnerability rediscovery.
• Protein design: Mythos 5.2 recorded an 85.1% score.

These figures come from Anthropic’s own testing, and differences in evaluation conditions may affect direct comparisons between models.

Anthropic demonstrates AI agents building an operating system

Beyond benchmarks, Anthropic claims Claude Fable 5.2 coordinated more than 1,000 AI agents over eight days to build a working operating system from scratch.

According to the company, the resulting operating system can boot on a laptop and includes device drivers and file storage.

Anthropic also claims the model designed three functional electronic devices, including a voice-assistant hub and a USB speakerphone.

In scientific research, the company says Fable 5.2 solved two previously open theoretical physics problems involving black holes. These demonstrations highlight the model’s potential for handling complex tasks that require multiple steps and extended reasoning.

Claude Fable 5.2 and Mythos 5.2 availability

Claude Fable 5.2 is available now through Claude, Claude Code, the Claude API, AWS, Google Cloud, and Microsoft Azure.

Developers can access the model using the claude-fable-5-2 model ID.

Meanwhile, Claude Mythos 5.2 is limited to verified organizations participating in Anthropic’s Cyber and Life Sciences Verification Programs. Enterprise customers can also access Mythos capabilities through Claude Security.

Anthropic is offering additional Fable usage to eligible paid subscribers during launch week.

Anthropic has also expanded Claude with personal finance tools and Claude Duet, an AI agent that works across apps while users sleep.