Claude Fable 5
Anthropic·
LLMsflagshipcloud
Overview
Anthropic's first publicly available Mythos-class model, priced at $10/M input and $50/M output. Launched alongside the gated Claude Mythos 5 (same base model with safeguards lifted for a small set of cyberdefenders and infrastructure providers), which is not publicly available and is not tracked here. A safety system routes a minority of sessions (~5% on average) to Claude Opus 4.8. Vendor-reported benchmarks: SWE-bench Pro 80.3, HLE 59.0 (no tools; 64.5 with tools), Terminal-Bench 2.1 88.0.
Capabilities and innovations
Autonomous long-horizon tasksVision inputAgentic workflowsSoftware engineeringFirst publicly available Mythos-class modelSafety routing to Claude Opus 4.8
Benchmarks
| Benchmark | Score | Source |
|---|---|---|
| GPQA Diamond | 93.18% | independent GPQA Diamond 93,18%, measured by Vals AI. Anthropic reports no GPQA value in the launch post; Artificial Analysis measures 92,63% at max effort. |
| Humanity’s Last Exam (no tools) | 59% | vendor HLE 59.0 without tools, 64.5 with tools. No-tools figure recorded for cross-model comparability. |
| SWE-bench Pro | 80.3% | vendor SWE-bench Pro 80.3 — Anthropic launch announcement. |
| MMLU-Pro | 91.5% | third party MMLU Pro 91.5 — rank 1 on the vals.ai MMLU-Pro leaderboard. Not reported in Anthropic's launch table. |
| Terminal-Bench 2.x | 80.52% | independent Terminal-Bench 2.1 80.52%, measured by Vals AI with the Terminus 2 harness (accessed 2026-09-23). Vals AI is the reference harness for this field so values compare across vendors; vendor and other third-party figures stay in the notes below. Replaces 88 (vendor). |
| Terminal-Bench 3.0 | 34% | independent Terminal-Bench 3.0 34% (± 3.4, 95% CI), official leaderboard maintained by the Terminal-Bench team (Stanford / Harbor / Laude Institute), max effort, agent harness: Claude Code (accessed 2026-09-23). Do not compare against the terminal_bench column: 3.0 is a harder generation on a different scale. |
| Terminal-Bench 4.0 | 44.5% | independent Terminal-Bench 4.0 44.5% (± 3.8, 95% CI), official leaderboard maintained by the Terminal-Bench team (Stanford / Harbor / Laude Institute), max effort, agent harness: Claude Code (accessed 2026-09-23). Vendor-reported: 42 (https://www.anthropic.com/claude-fable-and-mythos-5-1). 4.0 is its own generation and not comparable with the 2.x or 3.0 fields. |
- SWE-bench Verified: not reported by the vendor — Anthropic reports SWE-bench Pro, not Verified, for this release.
- MMLU: not reported by the vendor
- MMMU-Pro: not reported by the vendor
Reliability
- Agentic tool use (τ-bench)
- 89.2% (higher is better)third party
- tau3_banking
- 38.1 ()third party
- omniscience_index
- 43.3 ()independent
Links
More from Anthropic
Data curated by AI Model Timeline. See the methodology for admission criteria, benchmark eras and source priority.