Models › Text › Claude Fable 5
Anthropic · Launched June 2026 for autonomous work
Claude Fable 5 is Anthropic’s Mythos-class model for autonomous work
Starting provider API rate; context tiers apply
See what Fable 5 does best, then review its limits, benchmarks, and pricing below
Claude Fable 5 is Anthropic’s Mythos-class model for autonomous work
Every Fable 5 result below comes from a named source and is labeled by protocol, so each score reflects how it was measured
| Benchmark | Score | Evaluation details | Source |
|---|---|---|---|
Humanity's Last Exam A 2,500-question expert-level benchmark spanning dozens of academic fields. Tool-assisted and no-tools results are separate protocols. |
53% Accuracy |
Version: Humanity's Last Exam • Split: Text-only questions • Harness: Artificial Analysis independent evaluation • Fallback: ~9% of tasks routed to a safety fallback model • Reasoning effort: max | Methodology Artificial Analysis Independently reproduced |
GPQA Diamond The highest-quality subset of Graduate-Level Google-Proof Q&A, designed to test expert-level scientific reasoning in biology, physics, and chemistry. |
93.2% Accuracy |
Version: GPQA Diamond 198 • Split: 198 Diamond questions • Harness: Artificial Analysis independent evaluation • Reasoning effort: max | Methodology Artificial Analysis Independently reproduced |
SWE-bench Verified A 500-task benchmark of real GitHub issues, graded by whether the generated patch passes hidden unit tests. |
95.0% Pass@1 |
Version: SWE-bench Verified • Split: 500 verified tasks • Harness: Vals AI independent evaluation • Reasoning effort: max | Methodology Vals AI Independently reproduced |
GDPval-AA A knowledge-work benchmark scoring long, realistic professional tasks against expert-produced deliverables across dozens of occupations. |
1,932 Elo score |
Version: GDPval-AA • Harness: Artificial Analysis independent evaluation • Scoring: Pairwise Elo comparison against expert-produced work • Reasoning effort: max | Methodology Artificial Analysis Independently reproduced |
AutomationBench A benchmark of multi-step business automation tasks that chain tool calls across simulated workplace systems. |
17.1% Task success rate |
Version: AutomationBench • Harness: Artificial Analysis independent evaluation • Reasoning effort: max | Methodology Artificial Analysis Independently reproduced |
Arena rankings capture how people vote between anonymous responses. Fable 5’s human preference standing is reported per leaderboard, since results can vary widely by task
Arena’s overall leaderboard places Claude Fable 5 among the top tracked models, with an 8.81% win rate (±1.25%) and a 1,506 Elo rating
Claude Fable 5 runs a full 1,000,000-token context window at one flat rate, with image input and adjustable reasoning effort, priced and detailed below
| Model | |
| Provider | Anthropic |
| Release date | Jun 9, 2026 |
| Catalog added | Sep 22, 2026 |
| Context window | 1,000,000 tokens |
| Maximum output | 128,000 tokens |
| Knowledge cutoff | 2026-01-31 |
| License | Proprietary |
| Capabilities and API | |
| Input types | Image, Text |
| Output types | Text |
| Reasoning | Supported (adaptive, effort-adjustable) |
| Provider endpoint tool parameters | Accepts tool parameters |
| Base provider API input | $10.00 / 1M tokens |
| Base provider cached input (read) | $1.00 / 1M tokens |
| Base provider API output | $50.00 / 1M tokens |
| Cache write (5 min / 1 hour) | $12.50 / 1M tokens • $20.00 / 1M tokens |
Provider token rates are context-dependent. The base rates apply below the listed prompt thresholds; the matching tier applies at or above each threshold. These are reference provider API rates, not Pyxa AI checkout charges — access in Pyxa AI follows the allowances of your Pyxa AI plan.
Each Fable 5 benchmark and specification links to its source, so you can check where every figure comes from
Discover where Claude Fable 5 performs best, from autonomous coding and research to long-horizon agentic automation
Put Claude Fable 5 to work on your coding, research, and long-running agentic tasks inside Pyxa AI