Anthropic launches Claude Fable 5.1 and says it’s up to 45 percent cheaper for agentic work

THE NEWS

What happened

Anthropic has launched Claude Fable 5.1 and Mythos 5.1, addressing customer concerns regarding cost, data privacy, and safety. The new models offer significant performance improvements over their predecessors while reducing costs by up to 45% for complex agentic tasks through optimized cached data usage. Key updates include enhanced safeguards that allow more biology-related questions, a shift toward storing enterprise data on customer cloud servers for complete privacy, and expanded availability of Fable 5.1 across all platforms.

CONTEXT

Why it matters

Anthropic has officially launched its latest AI models: Claude Fable 5.1 and Mythos 5.1. The company is tackling three major pain points: price, data retention, and safety. Key highlights for North American developers: - Cost Efficiency: Fable 5.1 costs up to 45% less for complex agentic tasks by optimizing cached data usage. - Privacy Upgrade: Enterprise users can soon store data on their own cloud servers for complete privacy. - Safety & Capabilities: New safeguards allow more biology questions without blocking basic queries, while still redirecting high-risk cybersecurity tasks to Opus models. Fable 5.1 is available everywhere; Mythos 5.1 is currently limited to Project Glasswing participants. Early impressions suggest it is the strongest coding model used so far, offering better token efficiency and natural conversation.

AT A GLANCE

Key facts

• Claude Fable 5.1 costs around 25% less typically and up to 45% less for complex agentic tasks compared to Fable 5.
• The cost reduction is achieved through reduced pricing on cached data that was already processed and stored.
• Mythos 5.1 launches exclusively in Project Glasswing, while Fable 5.1 is available on all platforms.
• Fable 5.1 features more precise safeguards that are less likely to block basic biology questions than the previous model.
• Enterprise Frontier Safeguards will store data on customer cloud servers instead of Anthropic's own infrastructure starting later this fall.
• Anthropic now allows Fable 5.1 for identifying software vulnerabilities but redirects cybersecurity tasks like penetration testing to Opus models.
• Early adopters report Fable 5.1 is the strongest coding model used, offering better token efficiency and natural language interaction.