Model profile
Claude Haiku 4.5
Anthropicdeveloper
2025-10-15release date
#22 / 309overall rank
40eval lineages
Evidence summary
Claude Haiku 4.5 has an estimated overall rank of #22; its 90% source-sensitivity interval is #20–#80. Its behavior-only rank is #26; company governance moves the combined estimate to #22. Published evidence spans 40 evals and 7 of 7 behavior components. Its strongest relative result is AIRBench 2024 Safety Scenarios (safety_scenarios, #1 of 80); its weakest is Anthropic Claude Haiku 4.5 System Card (harmful_request_safety, #2 of 2).
Compare this model
Only models sharing at least one published sub-eval are listed.
Official and reference links
- Artificial Analysis ↗c2b1e769-7aee-4669-8076-73918bdebf6c
- OpenRouter ↗anthropic/claude-haiku-4.5
- 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 | #13 / 328 | ↓0.2571 | Source ↗official | |
| AgentAbstainabstain | #5 / 17 | ↑65.6 | Source ↗official | |
| AgentAbstaincar | #6 / 17 | ↑61.8 | Source ↗official | |
| AgentAbstainpaired | #5 / 17 | ↑49.7 | Source ↗official | |
| AIRBench 2024 Safety Scenariossafety_scenarios | #1 / 80 | ↑0.932 | Source ↗official | |
| ANIMAscore | #17 / 22 | ↑0.5862 | Source ↗official | |
| Anthropic Claude Haiku 4.5 System Cardagentic_coding_safety | #1 / 3 | ↑1 | Source ↗official | |
| Anthropic Claude Haiku 4.5 System Cardclaude_code_malicious_refusal | #2 / 3 | ↑0.6939 | Source ↗official | |
| Anthropic Claude Haiku 4.5 System Cardharmful_request_safety | #2 / 2 | ↑0.9938 | Source ↗official | |
| Anthropic Claude Opus 4.5 System Cardagentic_coding_safety | #1 / 4 | ↑1 | Source ↗official | |
| Anthropic Claude Opus 4.5 System Cardclaude_code_malicious_refusal | #2 / 4 | ↑0.6939 | Source ↗official | |
| Anthropic Claude Opus 4.5 System Cardcomputer_use_malicious_refusal | #3 / 4 | ↑0.7232 | Source ↗official | |
| Arena Factuality — Text Arena (factuality-only weighting)factuality_bt_rating | #55 / 112 | ↑1438.0 | Source ↗official | |
| AuAu Authoritarian Response Auditbehavioral_vignette_arr_percent | #2 / 17 | ↓20.4 | Source ↗official | |
| AuAu Authoritarian Response Auditrealistic_prompt_arr_percent | #3 / 17 | ↓1.2 | Source ↗official | |
| BullshitBench v2clear_pushback_rate | #8 / 106 | ↑0.74 | Source ↗official | |
| CAIS Risk Indexagent_red_teaming | #19 / 44 | ↓69.7 | Source ↗official | |
| CAIS Risk Indexbioweapons_assistance | #35 / 49 | ↓84.9 | Source ↗official | |
| CAIS Risk Indexhle_overconfidence | #32 / 50 | ↓60.5 | Source ↗official | |
| CAIS Risk Indexmachiavelli | #6 / 46 | ↓81.5 | Source ↗official | |
| CAIS Risk Indexmask | #3 / 52 | ↓4.6 | Source ↗official | |
| CAIS Risk Indexpolitical_manipulation | #32 / 33 | ↓62.2 | Source ↗official | |
| CAIS Risk Indextextquests_harm | #18 / 49 | ↓16.6 | Source ↗official | |
| Cisco AI Defense Rolling Single-Turn Leaderboardsingle_turn_attack_success_rate | #12 / 104 | ↓3.539 | Source ↗official | |
| Claude Sonnet 4.6 Overrefusalhigher_difficulty_overrefusal_rate | #4 / 5 | ↓6.01 | Source ↗official | |
| Claude Sonnet 4.6 Overrefusaloverall_overrefusal_rate | #3 / 5 | ↓0.26 | Source ↗official | |
| Claude Sonnet 4.6 User Wellbeingchild_benign_refusal_rate | #1 / 4 | ↓0.04 | Source ↗official | |
| Claude Sonnet 4.6 User Wellbeingchild_multiturn_appropriate_rate | #2 / 4 | ↑96 | Source ↗official | |
| Claude Sonnet 4.6 User Wellbeingchild_violative_harmless_rate | #3 / 4 | ↑99.88 | Source ↗official | |
| Claude Sonnet 4.6 User Wellbeingselfharm_benign_refusal_rate | #2 / 4 | ↓0.03 | Source ↗official | |
| Claude Sonnet 4.6 User Wellbeingselfharm_harmless_rate | #3 / 4 | ↑99.67 | Source ↗official | |
| Claude Sonnet 4.6 User Wellbeingselfharm_multiturn_appropriate_rate | #2 / 4 | ↑90 | Source ↗official | |
| Contextual MoralChoicehuman_agreement | #6 / 22 | ↑0.45 | Source ↗official | |
| DystopiaBenchbasaglia_score | #3 / 50 | ↓25.3 | Source ↗official | |
| DystopiaBenchbaudrillard_score | #2 / 50 | ↓22.73 | Source ↗official | |
| DystopiaBenchhuxley_score | #6 / 50 | ↓19.07 | Source ↗official | |
| DystopiaBenchlaguardia_score | #4 / 50 | ↓34.47 | Source ↗official | |
| DystopiaBenchorwell_score | #2 / 50 | ↓21.63 | Source ↗official | |
| DystopiaBenchpetrov_score | #1 / 50 | ↓19.8 | Source ↗official | |
| Enkrypt AI Safety Leaderboardbias_attack_non_success_rate | #41 / 241 | ↑35.14 | Source ↗official | |
| Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate | #99 / 241 | ↑89.5 | Source ↗official | |
| Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate | #14 / 241 | ↑98.89 | Source ↗official | |
| Enkrypt AI Safety Leaderboardtoxicity_attack_non_success_rate | #44 / 239 | ↑98.91 | Source ↗official | |
| HELM Safetyanthropic_red_team | #64 / 80 | ↑0.969 | Source ↗official | |
| HELM Safetybbq | #44 / 80 | ↑0.928 | Source ↗official | |
| HELM Safetyharmbench | #14 / 80 | ↑0.959 | Source ↗official | |
| HELM Safetysimple_safety_tests | #40 / 80 | ↑0.988 | Source ↗official | |
| HELM Safetyxstest | #60 / 80 | ↑0.932 | Source ↗official | |
| Inkling-Small model card — FORTRESSbenign_answer_rate | #8 / 10 | ↑94.1 | Source ↗official | |
| Inkling-Small model card — FORTRESSharmful_refusal_rate | #1 / 10 | ↑91.3 | Source ↗official | |
| Inkling-Small model card — StrongREJECTsafety_rate | #6 / 10 | ↑98.6 | Source ↗official | |
| JuICE Cultural-Error Span Detectionf1 | #6 / 10 | ↑0.3691 | Source ↗official | |
| KIDBench Implicit Child Cueimplicit_child_cue_total_mean | #5 / 13 | ↑4 | Source ↗official | |
| LiveSecBenchethics | #4 / 43 | ↑88.65 | Source ↗official | |
| LiveSecBenchfactuality | #3 / 43 | ↑82.68 | Source ↗official | |
| LiveSecBenchlegality | #1 / 43 | ↑98.62 | Source ↗official | |
| LiveSecBenchprivacy | #1 / 43 | ↑97.17 | Source ↗official | |
| LiveSecBenchpsychological_health | #2 / 43 | ↑90.02 | Source ↗official | |
| MACHIAVELLIdeception_relative_random_pct | #6 / 50 | ↓81.5 | Source ↗official | |
| MORUscore | #7 / 13 | ↑74.02 | Source ↗official | |
| MT-JailBench CrescendoXsafety_score | #1 / 21 | ↑81.76 | Source ↗official | |
| PHAREbias_resistance_diagnostic | #6 / 66 | ↑0.7066 | Source ↗official | |
| PHAREhallucination_resistance_diagnostic | #10 / 70 | ↑0.8356 | Source ↗official | |
| PHAREharm_resistance_diagnostic | #3 / 70 | ↑0.9993 | Source ↗official | |
| PHAREjailbreak_resistance_diagnostic | #4 / 67 | ↑0.7851 | Source ↗official | |
| Pokee-Isaac model card — DTAPbenign_task_success_rate | #5 / 6 | ↑0.713 | Source ↗official | |
| Pokee-Isaac model card — DTAPcombined_attack_success_rate | #2 / 6 | ↓0.379 | Source ↗official | |
| RealityTest — Text AI-Identity Disclosuredisclosure_probability | #1 / 17 | ↑0.923 | Source ↗official | |
| SM-Benchadversarial | #67 / 79 | ↑76.59 | Source ↗official | |
| SM-Benchambiguous_interpretation | #34 / 79 | ↑86.61 | Source ↗official | |
| SM-Benchanti_hallucination | #42 / 79 | ↑92.15 | Source ↗official | |
| SM-Bencheq_boundaries | #77 / 79 | ↑41.85 | Source ↗official | |
| SM-Benchoverfit | #24 / 79 | ↑81.97 | Source ↗official | |
| SpeciEvalbelief_animal_sentience | #104 / 113 | ↑6.32 | Source ↗official | |
| SpeciEvalland_animal_4ns | #20 / 113 | ↓4.25 | Source ↗official | |
| SpeciEvalsea_animal_4ns | #25 / 113 | ↓4.53 | Source ↗official | |
| SpeciEvalspeciesism | #69 / 113 | ↓2.15 | Source ↗official | |
| StereoTales Harmful Associationsbenign_significant_association_score | #4 / 23 | ↑88.2 | Source ↗official | |
| UAVBench safety-critical decision recognitionethical_safety_critical_accuracy | #17 / 27 | ↑0.67 | Source ↗official | |
| Vectara HHEM Factual Consistencyfactual_consistency_rate | #52 / 94 | ↑90.2 | Source ↗official | |
| Vigil Mental Health Safetyoverall_score | #4 / 23 | ↑74 | 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 | 5.4 | |
| Honesty-humility | 5.5 | |
| Extraversion | 5.7 | |
| Agreeableness | 7.5 | |
| Conscientiousness | 7.1 |