Overview
Claude Fable 5.1 is a flagship model released by Anthropic on September 1, 2026, sharing the same underlying architecture as Claude Mythos 5.1, but with a safety configuration fully open to all users and enterprises. Officially positioned as the world's most advanced model for programming and knowledge work, it is a flagship update three months after the release of Fable 5 in June 2026. The model demonstrates significant improvements in coding, scientific reasoning, and long-horizon tasks, accompanied by a substantial reduction in cache read pricing, aiming to provide developers and researchers with a more efficient and cost-effective agentic work experience.
Key Features
- Flagship-level coding and knowledge work capabilities: Achieves 52.6% on Terminal-Bench-Science 0.1, 73.4% on CursorBench, and 55.8% on Terminal-Bench 4.0, all significantly improved over the previous generation Fable 5, especially excelling at fixing root causes rather than applying quick patches.
- Coherence in long-horizon multi-file, multi-tool sessions: Maintains coherence in complex multi-file, multi-tool coding sessions, avoiding shortcuts that compromise output quality, suitable for deep agentic workloads.
- 1M context and 128K output: Supports a 1 million token context window and a maximum output of 128,000 tokens, capable of handling extremely long codebases or scientific documents.
- 75% reduction in cache read pricing: Cache read price reduced from $1 to $0.25 per million tokens, lowering typical workload costs by approximately 25% compared to Fable 5, with highly agentic workloads seeing reductions up to about 45%.
- More precise cybersecurity guardrails: Compared to Fable 5, cybersecurity guardrails are more precise, reducing safety interventions per session for Claude Code users by about 60%, while still allowing defensive vulnerability discovery, though exploit generation and penetration testing remain restricted.
- Distillation resistance and safety enhancements: No Critical-level jailbreaks found; new mechanisms added to counter distillation attacks; in new API accounts, previous thought records can no longer be manually edited in multi-turn conversations.
Use Cases
- Complex software fault diagnosis: Investment firm Millennium used Fable 5.1 to pinpoint a rare software crash that had puzzled engineers for years, ultimately tracing it to an external vendor library.
- Planetary science research: Participated in data processing for NASA's Magellan mission, training neural networks to build higher-resolution elevation maps covering about one-third of Venus's surface, improving detail from 10-20 kilometers to 2-3 kilometers.
- Long-horizon coding and codebase maintenance: Maintains coherence in multi-file, multi-tool sessions, suitable for large-scale project refactoring and root cause fixes.
- Knowledge-intensive workflows: Utilizes 1M context to process extremely long documents, scientific papers, or complex technical manuals, supporting deep analysis and synthesis.
Pros
- Outperforms on multiple official benchmarks, especially significantly ahead of the previous generation in scientific coding and terminal tasks.
- Cache read pricing significantly reduced, making long-running costs more economical.
- Good coherence in long-horizon tasks, reducing output quality loss.
- Fewer safety interventions with more precise guardrails, while maintaining defensive safety capabilities.
- Supports 1M context and 128K output, suitable for large-scale data processing.
- Public version available, open to all users and enterprises, easy to access.
Pricing
Standard input and output prices are $10 and $50 per million tokens, respectively, unchanged from Fable 5; cache read price reduced to $0.25 per million tokens. For specific pricing, refer to the official website.
Summary
Claude Fable 5.1 is a flagship model released by Anthropic in September 2026, sharing the same origin as Mythos 5.1, targeting the public market. It performs excellently on coding and scientific task benchmarks, with significantly reduced cache costs, and enhanced safety and distillation resistance. Suitable for complex coding, scientific research, and long-horizon agentic workloads.
Version History
- Claude Fable 5.1 launched alongside Mythos 5.1 (2026-09-01): Anthropic released Claude Fable 5.1, billed as the world most advanced model for coding and knowledge work: 52.6% on Terminal-Bench-Science 0.1 (Fable 5: 24.7%), 73.4% on CursorBench and 55.8% on Terminal-Bench 4.0. Cache-read pricing dropped 75% to $0.25 per million tokens, cutting typical workload costs by about 25% (up to ~45% for highly agentic workloads). Standard pricing stays at $10/$50 per million tokens with a 1M context and 128K output. More precise cyber safeguards cut safety interventions per Claude Code session by about 60%.