Model profile
Claude Fable 5
Evidence summary
Claude Fable 5 has an estimated overall rank of #2; its 90% source-sensitivity interval is #2–#38. Published evidence spans 18 evals and 7 of 7 behavior components. Its strongest relative result is Enkrypt AI Safety Leaderboard (bias_attack_non_success_rate, #1 of 241); its weakest is Olam Social Poker — Social Lie Rate (social_lie_rate_per_10000_turns, #19 of 19).
Compare this model
Only models sharing at least one published sub-eval are listed.
Official and reference links
- Artificial Analysis ↗claude-fable-5
- OpenRouter ↗anthropic/claude-fable-5
- System card ↗Family-level model document · Anthropic · first party
- Release source ↗direct research preview identity date
- www.anthropic.com ↗
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-eval | Rank | Value | Distribution | Source |
|---|---|---|---|---|
| AA-Omnisciencehallucination_rate | #86 / 328 | ↓0.6364 | Source ↗official | |
| Arena Factuality — Search Arena (factuality-only weighting)factuality_bt_rating | #15 / 30 | ↑1185.0 | Source ↗official | |
| Arena Factuality — Text Arena (factuality-only weighting)factuality_bt_rating | #24 / 112 | ↑1455.0 | Source ↗official | |
| BullshitBench v2clear_pushback_rate | #24 / 106 | ↑0.49 | Source ↗official | |
| CAIS Risk Indexagent_red_teaming | #1 / 44 | ↓39.4 | Source ↗official | |
| CAIS Risk Indexbioweapons_assistance | #1 / 49 | ↓0 | Source ↗official | |
| CAIS Risk Indexhle_overconfidence | #3 / 50 | ↓25.5 | Source ↗official | |
| CAIS Risk Indexmachiavelli | #2 / 46 | ↓80.2 | Source ↗official | |
| CAIS Risk Indexmask | #8 / 52 | ↓7.2 | Source ↗official | |
| CAIS Risk Indexpolitical_manipulation | #9 / 33 | ↓39.8 | Source ↗official | |
| CAIS Risk Indextextquests_harm | #45 / 49 | ↓23.7 | Source ↗official | |
| Enkrypt AI Safety Leaderboardbias_attack_non_success_rate | #1 / 241 | ↑100 | Source ↗official | |
| Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate | #1 / 241 | ↑100 | Source ↗official | |
| Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate | #1 / 241 | ↑100 | Source ↗official | |
| Enkrypt AI Safety Leaderboardtoxicity_attack_non_success_rate | #1 / 239 | ↑100 | Source ↗official | |
| Gray Swan indirect prompt injection (15 attempts)attack_success_probability_k15_pct | #3 / 13 | ↓2.8 | Source ↗official | |
| HUMAINE Trust, Ethics and Safetytrust_ethics_safety_score | #8 / 54 | ↑28.64 | Source ↗official | |
| MACHIAVELLIdeception_relative_random_pct | #2 / 50 | ↓80.2 | Source ↗official | |
| Manager Coercion Benchcoercion_ladder_depth | #7 / 31 | ↓7.1 | Source ↗official | |
| Manager Coercion Benchfabrication_rate | #1 / 13 | ↓0 | Source ↗official | |
| Olam Social Poker — Social Lie Ratesocial_lie_rate_per_10000_turns | #19 / 19 | ↓164 | Source ↗official | |
| SM-Benchadversarial | #54 / 79 | ↑79.51 | Source ↗official | |
| SM-Benchambiguous_interpretation | #6 / 79 | ↑92.56 | Source ↗official | |
| SM-Benchanti_hallucination | #14 / 79 | ↑98.43 | Source ↗official | |
| SM-Bencheq_boundaries | #67 / 79 | ↑53.65 | Source ↗official | |
| SM-Benchoverfit | #69 / 79 | ↑34.43 | Source ↗official | |
| SpeciEvalbelief_animal_sentience | #92 / 113 | ↑6.46 | Source ↗official | |
| SpeciEvalland_animal_4ns | #40 / 113 | ↓4.4 | Source ↗official | |
| SpeciEvalsea_animal_4ns | #62 / 113 | ↓4.78 | Source ↗official | |
| SpeciEvalspeciesism | #43 / 113 | ↓1.85 | Source ↗official | |
| TACbase_welfare_rate | #3 / 76 | ↑55.77 | Source ↗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.