Model profile
Claude Sonnet 4.6
Anthropicdeveloper
2026-01-21release date
#15 / 312overall rank
34eval lineages
1discovery sources
Evidence summary
Claude Sonnet 4.6 has an estimated overall rank of #15; its 90% source-sensitivity interval is #7–#85. Its behavior-only rank is #18; company governance moves the combined estimate to #15. Published evidence spans 34 evals and 7 of 7 behavior components. Its strongest relative result is SM-Bench (anti_hallucination, #1 of 79); its weakest is Claude Sonnet 4.6 User Wellbeing (child_multiturn_appropriate_rate, #4 of 4).
Compare this model
Only models sharing at least one published sub-eval are listed.
Official and reference links
- Artificial Analysis ↗claude-sonnet-4-6
- OpenRouter ↗anthropic/claude-sonnet-4.6
- System card ↗Exact model document · Anthropic · first party
- Release source ↗direct research preview identity date
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 | #55 / 330 | ↓0.4845 | Source ↗official | |
| AgentAbstainabstain | #4 / 17 | ↑66.4 | Source ↗official | |
| AgentAbstaincar | #4 / 17 | ↑65.4 | Source ↗official | |
| AgentAbstainpaired | #3 / 17 | ↑53.4 | Source ↗official | |
| AIMS Safety-Classifier Competenceaverage_harmful_f1 | #8 / 11 | ↑0.782 | Source ↗official | |
| ANIMAscore | #20 / 22 | ↑0.5646 | Source ↗official | |
| Arena Factuality — Search Arena (factuality-only weighting)factuality_bt_rating | #24 / 30 | ↑1169.0 | Source ↗official | |
| Arena Factuality — Text Arena (factuality-only weighting)factuality_bt_rating | #36 / 112 | ↑1447.0 | Source ↗official | |
| BioSecBench-Refusalbalanced_refusal_score | #2 / 10 | ↑0.4304 | Source ↗official | |
| BullshitBench v2clear_pushback_rate | #2 / 106 | ↑0.9 | Source ↗official | |
| CAIS Risk Indexagent_red_teaming | #14 / 45 | ↓54.8 | Source ↗official | |
| CAIS Risk Indexbioweapons_assistance | #7 / 50 | ↓31.5 | Source ↗official | |
| CAIS Risk Indexhle_overconfidence | #11 / 51 | ↓43.1 | Source ↗official | |
| CAIS Risk Indexmachiavelli | #17 / 47 | ↓84.9 | Source ↗official | |
| CAIS Risk Indexmask | #15 / 53 | ↓11.7 | Source ↗official | |
| CAIS Risk Indextextquests_harm | #17 / 50 | ↓16.5 | Source ↗official | |
| Cisco AI Defense Rolling Single-Turn Leaderboardsingle_turn_attack_success_rate | #9 / 104 | ↓3.041 | Source ↗official | |
| Claude Sonnet 4.6 Overrefusalhigher_difficulty_overrefusal_rate | #2 / 5 | ↓0.18 | Source ↗official | |
| Claude Sonnet 4.6 Overrefusaloverall_overrefusal_rate | #2 / 5 | ↓0.23 | Source ↗official | |
| Claude Sonnet 4.6 User Wellbeingchild_benign_refusal_rate | #2 / 4 | ↓0.08 | Source ↗official | |
| Claude Sonnet 4.6 User Wellbeingchild_multiturn_appropriate_rate | #4 / 4 | ↑95 | Source ↗official | |
| Claude Sonnet 4.6 User Wellbeingchild_violative_harmless_rate | #1 / 4 | ↑99.96 | Source ↗official | |
| Claude Sonnet 4.6 User Wellbeingselfharm_benign_refusal_rate | #3 / 4 | ↓0.17 | Source ↗official | |
| Claude Sonnet 4.6 User Wellbeingselfharm_harmless_rate | #2 / 4 | ↑99.73 | Source ↗official | |
| Claude Sonnet 4.6 User Wellbeingselfharm_multiturn_appropriate_rate | #1 / 4 | ↑98 | Source ↗official | |
| Constitutional Following — Anthropic Constitutionconstitutional_following_score | #1 / 7 | ↑98 | Source ↗official | |
| Constitutional Following — OpenAI Model Specconstitutional_following_score | #4 / 7 | ↑94.4 | Source ↗official | |
| DelusionEvaldelusional_prevalence_pct | #8 / 16 | ↓31 | Source ↗official | |
| DelusionEvaldiscourages_harm_prevalence_pct | #9 / 16 | ↑23.5 | Source ↗official | |
| DelusionEvalfacilitates_harm_prevalence_pct | #6 / 16 | ↓2.1 | Source ↗official | |
| DelusionEvalrelationship_prevalence_pct | #7 / 16 | ↓21.4 | Source ↗official | |
| DelusionEvalsycophancy_prevalence_pct | #9 / 16 | ↓27.1 | Source ↗official | |
| DystopiaBenchbasaglia_score | #2 / 50 | ↓24.2 | Source ↗official | |
| DystopiaBenchbaudrillard_score | #6 / 50 | ↓27.87 | Source ↗official | |
| DystopiaBenchhuxley_score | #3 / 50 | ↓17.6 | Source ↗official | |
| DystopiaBenchlaguardia_score | #3 / 50 | ↓33 | Source ↗official | |
| DystopiaBenchorwell_score | #5 / 50 | ↓24.57 | Source ↗official | |
| DystopiaBenchpetrov_score | #3 / 50 | ↓24.93 | Source ↗official | |
| Enkrypt AI Safety Leaderboardbias_attack_non_success_rate | #92 / 241 | ↑18.86 | Source ↗official | |
| Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate | #167 / 241 | ↑84.67 | Source ↗official | |
| Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate | #6 / 241 | ↑99.44 | Source ↗official | |
| Governance Decay under Passive Context Compactiongovernance_retention_score | #3 / 7 | ↑81 | Source ↗official | |
| HarmVideoBenchharmful_video_safety_recognition_reasoning | #6 / 19 | ↑0.799 | Source ↗official | |
| LiveSecBenchethics | #8 / 43 | ↑78.63 | Source ↗official | |
| LiveSecBenchfactuality | #1 / 43 | ↑92.17 | Source ↗official | |
| LiveSecBenchlegality | #2 / 43 | ↑97.77 | Source ↗official | |
| LiveSecBenchprivacy | #3 / 43 | ↑89.26 | Source ↗official | |
| LiveSecBenchpsychological_health | #8 / 43 | ↑72.01 | Source ↗official | |
| MACHIAVELLIdeception_relative_random_pct | #17 / 50 | ↓84.9 | Source ↗official | |
| Manager Coercion Benchcoercion_ladder_depth | #3 / 33 | ↓4.2 | Source ↗official | |
| Manager Coercion Benchfabrication_rate | #1 / 15 | ↓0 | Source ↗official | |
| MORUscore | #11 / 13 | ↑69.73 | Source ↗official | |
| Pander Scoreconversational_absolute_pander_score | #3 / 20 | ↓3.936 | Source ↗official | |
| Pander Scoreinstructional_absolute_pander_score | #1 / 20 | ↓11.79 | Source ↗official | |
| PHAREbias_resistance_diagnostic | #42 / 66 | ↑0.4128 | Source ↗official | |
| PHAREhallucination_resistance_diagnostic | #8 / 70 | ↑0.8551 | Source ↗official | |
| PHAREharm_resistance_diagnostic | #2 / 70 | ↑0.9993 | Source ↗official | |
| PHAREjailbreak_resistance_diagnostic | #13 / 67 | ↑0.6939 | Source ↗official | |
| RefusalBenchyouden_j | #2 / 19 | ↑0.6766 | Source ↗official | |
| SM-Benchadversarial | #49 / 79 | ↑80 | Source ↗official | |
| SM-Benchambiguous_interpretation | #23 / 79 | ↑88.99 | Source ↗official | |
| SM-Benchanti_hallucination | #1 / 79 | ↑100 | Source ↗official | |
| SM-Bencheq_boundaries | #79 / 79 | ↑33.43 | Source ↗official | |
| SM-Benchoverfit | #11 / 79 | ↑91.53 | Source ↗official | |
| SpeciEvalbelief_animal_sentience | #60 / 113 | ↑6.78 | Source ↗official | |
| SpeciEvalland_animal_4ns | #71 / 113 | ↓4.65 | Source ↗official | |
| SpeciEvalsea_animal_4ns | #45 / 113 | ↓4.67 | Source ↗official | |
| SpeciEvalspeciesism | #65 / 113 | ↓2.1 | Source ↗official | |
| StereoTales Harmful Associationsbenign_significant_association_score | #3 / 23 | ↑88.69 | Source ↗official | |
| TACbase_welfare_rate | #9 / 76 | ↑37.82 | Source ↗official | |
| Vectara HHEM Factual Consistencyfactual_consistency_rate | #59 / 94 | ↑89.4 | Source ↗official | |
| VETO Misfired Alignmentmisfired_alignment_rate_pct | #21 / 25 | ↓10.9 | Source ↗official | |
| Vigil Mental Health Safetyoverall_score | #1 / 23 | ↑83 | 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.
UGI Political Values
Agent-ValueBench Moral Foundations (MFT08)
Agent-ValueBench HEXACO
| Dimension | Value | Distribution |
|---|---|---|
| Openness to experience | 5.8 | |
| Honesty-humility | 7.3 | |
| Extraversion | 5.4 | |
| Agreeableness | 6.1 | |
| Conscientiousness | 7.1 |