Anthropic Launches Claude Fable 5 and Claude Mythos 5 Models
Anthropic has released the Fable 5 and Mythos 5 models, focusing on enhanced reasoning capabilities and expanded agentic workflows. These models aim to improve accuracy in complex multi-step tasks for developers.
Impact: Medium
Why it matters
Evaluate these new models for your existing agentic pipelines to see if they reduce hallucination rates in complex coding tasks.
TL;DR
- 01Enhanced reasoning for multi-step agent tasks
- 02Improved state retention during long coding sessions
- 03More granular control over model output constraints
Key facts
- Input Pricing
- $10 per million tokens
- Output Pricing
- $50 per million tokens
- Safeguard False Positives
- < 5% of sessions
- Stripe Migration Scale
- 50-million-line Ruby codebase
- Drug Design Acceleration
- approx. 10x faster
A Leap in Autonomous Workflows
Claude Fable 5 and Mythos 5 introduce advanced capabilities for long-running autonomous tasks. During testing, Stripe used Fable 5 to migrate a 50-million-line Ruby codebase in just one day—a task that would normally take an entire team over two months. Fable 5 also beat *Pokémon FireRed* using a minimal, vision-only harness, proving its state-of-the-art vision processing.
Advanced Scientific and Biotech Capabilities
Claude Mythos 5 is designed for specialized cyberdefense and scientific workflows. In drug design, internal protein design experts used Mythos 5 to accelerate aspects of the process by approximately 10x, successfully designing candidates for 9 out of 14 protein targets. In genomics, the model autonomously trained a custom ML model that was 100 times smaller than a model published in the journal *Science*, yet outperformed it.
Conservative Safeguards and Pricing
These models are priced at $10 per million input tokens and $50 per million output tokens, which is less than half the price of Claude Mythos Preview. To ensure safety, Fable 5 features conservative safeguards that trigger in less than 5% of sessions, routing sensitive queries to Claude Opus 4.8.
✓ When to use
- When building complex autonomous agent loops requiring high steering precision and persistent state.
- When processing heavy visual information, such as recreating frontend interfaces from screenshots.
- When conducting frontier-level scientific research in genomics or molecular biology.
✕ When NOT to use
- When absolute cost minimization is required, as cheaper legacy models may suffice.
- When developers cannot tolerate up to 5% of requests falling back to Opus 4.8 due to safeguards.
What to do today
- Update your API configuration to point to the new model identifiers
- Run your existing agent regression test suite against Fable 5
What the community says
“Check your /memory”
Sources