Skip to content

Anthropic

Claude Fable 5

First generally available “Mythos-class” model, and the top general model on OSWorld-Verified (86.0%). Access was suspended on 12 June 2026 under a US export-control directive and restored on 1 July.

Announced
Access
Public

Benchmark scores

TestScoreBest on the trackerHow it was measured
Epoch Capabilities Indexmany tests combined into one score164 pointsGPT-6.2 Astra 174 pointsIndependentlisted as "Claude Fable 5", CI 161.83 to 167.02Run 9 Nov 2026
Remote Labor Indexpaid freelance projects16%Claude Fable 5.2 41%IndependentFable 5Run 1 Jul 2026
FrontierSWE V2very long engineering jobs47%Claude Fable 5.2 70%Independentproximus harness, max effort, mean@5 over 34 tasks, 20-hour budget (best@5 61.8)
Terminal-Bench 4.1long command-line tasks45%Claude Opus 5.5 66%IndependentClaude Code harness, max effort, 330 trials, ±3.83 (95% CI)
MirrorCoderebuilding whole programs64%Claude Opus 5.5 79%Independenthigh reasoning effort, MirrorCode (ML, +Private, 2L): 15 programs x 2 languages, 3 runs each, 10B-token / 7-day budget, ±10.4 SERun 10 Aug 2026
APEX-Agentsprofessional office tasks64%Gemini 4 Carbon 84%Independentpass@1, default effort; listed as "Fable 5"
FrontierMath T1–3hard maths problems87%Claude Fable 5.2 95%Independentmax reasoning effort (claude-fable-5_max)Run 9 Jun 2026
OSWorld-Verifiedusing a computerPeople (original task set): 72%86%Top scoreIndependentplain model, max 100 steps, 1M-context variant ("claude-fable-5[1m]", 307.73 / 358 tasks)Run 1 Aug 2026
GPQA DiamondPhD-level science questionsPhD experts: 70%Saturated86%Gemini 4 Argon 96%Independentmax reasoning effort (claude-fable-5_max)Run 6 Aug 2026
FrontierMath T4research-level maths problemsSaturated90%GPT-6.1 Sol 100%Independentmax reasoning effort (claude-fable-5_max)Run 9 Jun 2026
ARC-AGI-2visual puzzles, model aloneAverage person: 60%Saturated89%GPT-6 Astra 95%IndependentARC Prize config "Claude Fable 5 (Max)" (anthropic-claude-fable-5-max)
OTIS Mock AIMEcompetition mathsSaturated100%Top scoreIndependenthigh reasoning effort (claude-fable-5_high)Run 6 Aug 2026
METR time horizon (TH1.1 + TH2)task length an AI agent finishes half the time16 hGPT-6 Astra 1.7 daysIndependentPooled TH1.1 + TH2 suites; 95% CI 534.8–1939.3 min

Compare every model on the same tests →

Also on these charts

Sources