Claude Fable 5
Claude Fable 5 by Anthropic: first publicly released Mythos-class model with adaptive thinking always on, multi-day autonomous agents, vision-assisted self-checking, and built-in safety classifiers.
Model Overview
Capabilities, design details, and architectural traits
Claude Fable 5 - The First Mythos-Class Model Built for Public Release
Claude Fable 5 is Anthropic's most capable widely released model and the first model derived from the Mythos class to be made generally available. Its core design principle is sustained, long-horizon agentic work - tasks that span days rather than minutes - with minimal human supervision. The longer and more complex the task, the larger its performance lead over previous Claude models.
What Makes Fable 5 Structurally Different
| Trait | What it means for Fable 5 specifically |
|---|---|
| Mythos-class with public safety classifiers | Shares the same underlying model as Claude Mythos 5, but ships with classifiers that automatically route flagged cybersecurity and biology queries to Opus 4.8 instead |
| Adaptive thinking, always on | The only thinking mode available on Fable 5 and Mythos 5; raw chain-of-thought is never returned - only a summarized or omitted version |
| Vision-assisted self-evaluation | Uses vision to check its own coding output against the original design or goal, not just to process input |
| Multi-day autonomous execution | Officially documented to work for days at a time inside agent harnesses like Claude Code or Claude Managed Agents |
| 30-day mandatory data retention | All Fable 5 traffic is subject to a 30-day retention requirement for safety monitoring, a policy unique to Mythos-class models |
| 1M token context, 128k output | Supports up to 128k output tokens per request alongside a 1 million token context window |
The Fable / Mythos Split
Fable 5 and Claude Mythos 5 share identical training and capabilities. The only documented difference is that Mythos 5 does not include the safety classifiers. Mythos 5 is not publicly available and is deployed exclusively through Project Glasswing in collaboration with the US government. Fable 5 exists specifically to bring Mythos-class capability to general use while keeping the highest-risk capabilities gated.
Benchmark Performance
Independent evaluations · Artificial Analysis
Accuracy & Capability Details
Independent evaluation data provided by Artificial Analysis. To view the latest benchmarks and full details, visit their official site.
Compare Models Side-by-Side
Evaluate specifications, pricing, and independent benchmark indices
| Model Details | AnthropicHost Claude Fable 5 | ||
|---|---|---|---|
| General Info | |||
| Provider | Anthropic | Anthropic | Anthropic |
| Release Date | June 9, 2026 | May 28, 2026 | April 16, 2026 |
| Knowledge Cutoff | — | — | — |
| Context & Limits | |||
| Context Window | 1M | 1M | 1M |
| Pricing (per 1M tokens) | |||
| Input Pricing | $10 | $5 Best Input Pricing | $5 Best Input Pricing |
| Output Pricing | $50 | $25 Best Output Pricing | $25 Best Output Pricing |
| Modalities | |||
| Inputs | textimagefile | textimagefile | textimagefile |
| Outputs | text | text | text |
| Benchmarks (0-100) | |||
| Intelligence Index | 59.9 Best Intelligence Index | 55.7 | 53.5 |
| Coding Index | 76.5 Best Coding Index | 74.3 | 73.6 |
| Agentic Index | 52.8 Best Agentic Index | 47.2 | 44.4 |
GPQA Benchmark
Graduate-level reasoning and expert Q&A evaluation.
Humanity's Last Exam
Extremely difficult logical reasoning and knowledge.
IFBench Instruction
Evaluation of strict instruction following.
Benchmark scores are independent evaluations sourced from Artificial Analysis. Intelligence, Coding, and Agentic indices reflect composite performance ratings (0–100).