Anthropic
Claude Fable 5
Anthropic's Mythos-class flagship — state-of-the-art on nearly every tested benchmark.
API onlytextimagefileREL 2026-06-09
Compare this model →Verified 2026-06-12
Overview
Claude Fable 5 is a Mythos-class model Anthropic has made safe for general use — state-of-the-art on nearly all tested benchmarks (#1 on SWE-bench Verified at 95%, top of the Artificial Analysis index at 65). Queries touching cybersecurity, bio/chem or distillation fall back to Opus 4.8 (under 5% of sessions). Its unrestricted sibling, Claude Mythos 5, is limited to vetted cyberdefenders via Project Glasswing. At $10/$50 per 1M tokens it's priced for the hardest work only.
Benchmark scores
Methodology →Intelligence65
Coding95
Reasoning93.2
Mathn/a
Sources and verification
- Coding: SWE-bench Verified · checked 2026-06-09
- Intelligence: Artificial Analysis Intelligence Index · checked 2026-06-12
- Reasoning: GPQA Diamond · checked 2026-06-09
- Catalogue and pricing: OpenRouter model record · checked 2026-06-12
Strengths and weaknesses
Strengths
- #1 on SWE-bench Verified (95%) and the AA intelligence index (65)
- Works autonomously for longer than any previous Claude
- State-of-the-art vision + long-context memory
Trade-offs
- $10/$50 per 1M — most expensive Anthropic tier
- Safety classifiers fall back to Opus 4.8 on some topics (<5% of sessions)
- Closed weights
Reach for this model when
The single hardest problems where you want the best available answer regardless of cost.
Run something like it locally
No local equivalent — closed weights, frontier-only capability.