Claude's Fable & Mythos 5.1: Next-Gen AI for Code & Knowledge

Anthropic's Leap in AI Capabilities

Anthropic's announcement of Claude Fable 5.1 and Claude Mythos 5.1 marks a notable step forward in large language model capabilities, particularly for coding and knowledge work. The claimed performance gains, especially the doubling of scores on Terminal-Bench-Science and significant improvements on Terminal-Bench 4.0, suggest a substantial leap in model intelligence and efficiency. The introduction of Enterprise Frontier Safeguards (EFS) is a critical move, addressing enterprise concerns around data privacy and security while maintaining state-of-the-art adversarial resistance. This directly targets a key adoption barrier for sensitive industries. Furthermore, the cost reductions, particularly the 75% less for cache reads and up to 45% reduction for agentic workloads, are highly practical and will make deploying these advanced models more economically viable.

The emphasis on research capabilities in Fable 5.1, hinting at contributions to scientific progress, is particularly exciting. This suggests a move beyond general-purpose AI towards specialized applications that can accelerate discovery. The reduction in false positives for cybersecurity and medical queries also indicates a maturing and more reliable AI system. However, the 'trusted access programs' for Claude Mythos 5.1 suggest that its most specialized capabilities are not yet broadly available, which might limit immediate widespread adoption for that specific model. The benchmarks, while impressive, are internal, and independent verification will be crucial for widespread industry acceptance. The exact technical underpinnings of the performance gains and cost reductions are not detailed, leaving room for speculation on the underlying architectural changes or training methodologies.

Key Points

• Introduction of Claude Fable 5.1 and Claude Mythos 5.1, positioned as the world's most advanced models for coding and knowledge work.
• Fable 5.1 shows significant benchmark improvements, doubling performance on Terminal-Bench-Science 0.1 and scoring 55.8% on Terminal-Bench 4.0.
• Enhanced cost-efficiency: Cache reads are 75% cheaper, reducing overall model costs by up to 45% for agentic workloads.
• Enterprise Frontier Safeguards (EFS) offer complete privacy (zero data retention) and state-of-the-art adversarial prevention for enterprise customers.
• Improved safeguards: Cybersecurity requests flagged 60% less often, and fallback rate reduced by 85% for basic biology/medical questions.
• Claude Fable 5.1 is generally available; Claude Mythos 5.1 is via trusted access programs.

---

📖 Source: [We’re introducing Claude Fable 5.1 and Claude Mythos 5.1.

They're the world’s most advanced models ...](https://x.com/claudeai/status/2094848572143407483)