Anthropic Launches Claude Fable 5: Frontier Model for Days-Long Autonomous Knowledge Work and Coding
Key Takeaways
- ▸Claude Fable 5 can work autonomously for multi-day periods, planning across stages, delegating to sub-agents, and validating its own work—unlocking a new class of long-horizon problems
- ▸State-of-the-art coding and vision capabilities: strongest results on CursorBench, can write and test code, evaluate outputs against goals, and understand complex documents with diagrams and tables
- ▸Priced at $10/M input tokens and $50/M output tokens with prompt caching discounts; available on Enterprise plan and Claude Platform plus major cloud providers (AWS, GCP, Azure)
Summary
Anthropic has announced Claude Fable 5, its fifth-generation flagship model designed to handle the most ambitious, long-running knowledge work and coding projects. The model represents a significant leap forward in autonomous reasoning and extended-horizon problem-solving, capable of sustaining complex work for days when deployed in agent harnesses like Claude Code or Claude Managed Agents. It can tackle multi-stage projects, research tasks, and large-scale code migrations with minimal human oversight.
Clause Fable 5 introduces several breakthrough capabilities: it autonomously writes and validates its own tests, uses vision to evaluate code against original designs, understands complex diagrams and documents, and excels at strategic reasoning and resource allocation. The model achieves state-of-the-art performance on CursorBench for coding tasks and delivers "senior research scientist-grade" reasoning for knowledge work. Early customer feedback highlights its ability to handle truly long-horizon problems previously out of reach for earlier models.
The model is available on Anthropic's consumption-based Enterprise plan and Claude Platform, as well as through AWS, Google Cloud, and Microsoft Azure. Pricing is set at $10 per million input tokens and $50 per million output tokens, with support for the existing 90% discount on cached input tokens. For US-only inference workloads, pricing is available at 1.1x the standard rate.
- Built-in safety mechanisms including automatic routing of sensitive cybersecurity and biology queries to Opus 4.8 at no additional charge
Editorial Opinion
Claude Fable 5 marks a meaningful step beyond task-completion toward genuine long-horizon autonomous agents. The ability to sustain complex work for days, write and validate tests without oversight, and engage in senior-level reasoning for both code and knowledge work suggests real maturation in AI capability. For developers and enterprises tackling genuinely ambitious problems—large migrations, novel research, multi-stage system design—this could unlock work that was previously impossible to automate. The vision integration and self-evaluation features demonstrate thoughtful engineering, and Anthropic's transparent safety approach (rerouting sensitive domains, publishing system cards) shows they're taking responsible deployment seriously.
