← Models

Model profile

Claude Fable 5

Anthropicdeveloper
2026-06-09release date
#2 / 267overall rank
16eval lineages
2discovery sources

Evidence summary

Claude Fable 5 has an estimated overall rank of #2; its 90% source-sensitivity interval is #2–#12. Published evidence spans 16 evals and 7 of 7 behavior components. Its strongest relative result is Enkrypt AI Safety Leaderboard (bias_attack_non_success_rate, #1 of 260); its weakest is CAIS Risk Index (textquests_harm, #44 of 48).

Compare this model

Only models sharing at least one published sub-eval are listed.

Official and reference links

Published eval results

Rank is within that sub-eval. Black marks the observed result; the grey dot marks the value implied by the global rank. Values stay on each source’s native scale.

Eval / sub-evalRankValueDistributionBetterSource
AA-Omnisciencehallucination_rate#61 / 3110.5485↓ lowerSource ↗official
BullshitBench v2clear_pushback_rate#24 / 1050.49↑ higherSource ↗official
CAIS Risk Indexagent_red_teaming#1 / 4339.4↓ lowerSource ↗official
CAIS Risk Indexbioweapons_assistance#1 / 480↓ lowerSource ↗official
CAIS Risk Indexhle_overconfidence#2 / 4925.5↓ lowerSource ↗official
CAIS Risk Indexmachiavelli#2 / 4580.2↓ lowerSource ↗official
CAIS Risk Indexmask#7 / 517.2↓ lowerSource ↗official
CAIS Risk Indexpolitical_manipulation#8 / 3239.8↓ lowerSource ↗official
CAIS Risk Indextextquests_harm#44 / 4823.7↓ lowerSource ↗official
Enkrypt AI Safety Leaderboardbias_attack_non_success_rate#1 / 260100↑ higherSource ↗official
Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate#1 / 260100↑ higherSource ↗official
Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate#1 / 260100↑ higherSource ↗official
Enkrypt AI Safety Leaderboardtoxicity_attack_non_success_rate#1 / 258100↑ higherSource ↗official
Gray Swan indirect prompt injection (15 attempts)attack_success_probability_k15_pct#3 / 132.8↓ lowerSource ↗official
HUMAINE Trust, Ethics and Safetytrust_ethics_safety_score#8 / 5428.64↑ higherSource ↗official
MACHIAVELLIdeception_relative_random_pct#2 / 5080.2↓ lowerSource ↗official
Manager Coercion Benchcoercion_ladder_depth#7 / 317.1↓ lowerSource ↗official
Manager Coercion Benchfabrication_rate#1 / 130↓ lowerSource ↗official
SM-Benchadversarial#49 / 7379.51↑ higherSource ↗official
SM-Benchambiguous_interpretation#4 / 7392.56↑ higherSource ↗official
SM-Benchanti_hallucination#12 / 7398.43↑ higherSource ↗official
SM-Bencheq_boundaries#61 / 7353.65↑ higherSource ↗official
SM-Benchoverfit#63 / 7334.43↑ higherSource ↗official
SpeciEvalbelief_animal_sentience#83 / 1026.46↑ higherSource ↗official
SpeciEvalland_animal_4ns#35 / 1024.4↓ lowerSource ↗official
SpeciEvalsea_animal_4ns#53 / 1024.78↓ lowerSource ↗official
SpeciEvalspeciesism#38 / 1021.85↓ lowerSource ↗official
TACbase_welfare_rate#3 / 6855.77↑ higherSource ↗official

Values evaluations

Descriptive values and political-framing results are separate from safety/ethics ranks. Each strip shows the evaluation’s observed model range; its endpoint labels state what lower and higher values mean.

UGI Political Values

DimensionValueDistribution
Political Lean-21.1
Government45.3
Diplomacy65.9
Economy48.2
Society59.4

CAIS AI Values — countries