MYTHOS is LIVE!!!!
Summary
Anthropic has publicly released Fable, a new "Mythos-class" large language model, on June 9, 2026, making it generally available with additional guard rails compared to the security-focused Mythos variant. This 10 trillion-parameter model demonstrates state-of-the-art performance across numerous benchmarks, including Aenta coding (80%), Agentic Coding Frontier Code Diamond (29.3%), and GDP val (1932). Priced at \$10 per million input tokens and \$50 per million output tokens, Fable excels in complex, long-horizon tasks, autonomously compressing months of engineering work into days, as seen in a 50 million-line Ruby codebase migration for Stripe. While more token-efficient algorithmically, its output is notably verbose and information-dense, often requiring simplification. The model also exhibits quirks like frequent clarifying questions and a slow initial processing phase before rapidly consuming tokens, especially when utilizing advanced features like Ultra Code workflows with parallel agents. A new 30-day data retention policy for business customer data is in place for safety, not model training.
Key takeaway
For AI Engineers and ML Directors integrating new frontier models, Fable presents a powerful, albeit expensive, option for highly complex, long-duration tasks. You should strategically route your most challenging problems to Fable, leveraging its autonomous capabilities and Ultra Code workflows, while utilizing more cost-effective models for simpler tasks. Be prepared for verbose outputs and an iterative clarification process, and start with lower effort settings to manage token consumption and cost effectively.
Key insights
Fable is a 10 trillion-parameter model excelling in complex, long-horizon tasks, but with notable verbosity and a unique interaction flow.
Principles
- Model capabilities scale with token usage, showing no clear limit.
- Information density in model output enhances efficiency.
- Strategic model routing optimizes cost for diverse tasks.
Method
The Ultra Code workflow employs a planning agent to delegate tasks to hundreds of parallel sub-agents, optimizing complex problem-solving.
In practice
- Start Fable tasks with lowest effort settings.
- Route complex, high-value tasks to Fable.
- Expect verbose, information-dense model outputs.
Topics
- Anthropic Fable
- Large Language Models
- AI Benchmarks
- Agentic Workflows
- Model Pricing
- Software Automation
Best for: CTO, VP of Engineering/Data, AI Architect, AI Engineer, Machine Learning Engineer, Director of AI/ML
Related on AIssential
Editorial summary, takeaway, and curation by AIssential. Original article published by Matthew Berman.