Model profile
Claude Opus 4.1
Evidence summary
Claude Opus 4.1 has an estimated overall rank of #36; its 90% source-sensitivity interval is #21–#104. Its behavior-only rank is #43; company governance moves the combined estimate to #36. Published evidence spans 15 evals and 6 of 7 behavior components. Its strongest relative result is Cisco AI Defense Rolling Single-Turn Leaderboard (single_turn_attack_success_rate, #2 of 104); its weakest is Anthropic Claude Opus 4.1 System Card Addendum (benign_request_refusal, #2 of 2).
Compare this model
Only models sharing at least one published sub-eval are listed.
Official and reference links
- OpenRouter ↗anthropic/claude-opus-4.1
- 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 |
|---|---|---|---|---|
| AgentDrive Safety Compliancescr | #10 / 48 | ↑93.75 | Source ↗official | |
| Anthropic Claude Opus 4.1 System Card Addendumbbq_disambiguated_accuracy | #2 / 2 | ↑0.907 | Source ↗official | |
| Anthropic Claude Opus 4.1 System Card Addendumbenign_request_refusal | #2 / 2 | ↓0.0008 | Source ↗official | |
| Anthropic Claude Opus 4.1 System Card Addendumharmful_request_safety | #1 / 2 | ↑0.9876 | Source ↗official | |
| Anthropic Claude Opus 4.5 System Cardagentic_coding_safety | #4 / 4 | ↑0.96 | Source ↗official | |
| Anthropic Claude Opus 4.5 System Cardclaude_code_malicious_refusal | #4 / 4 | ↑0.4816 | Source ↗official | |
| Anthropic Claude Opus 4.5 System Cardcomputer_use_malicious_refusal | #4 / 4 | ↑0.6696 | Source ↗official | |
| Anthropic Claude Opus 4.5 System Cardharmful_request_safety | #2 / 2 | ↑0.9914 | Source ↗official | |
| Anthropic Claude Sonnet 4.5 System Cardbenign_request_refusal | #2 / 2 | ↓0.0008 | Source ↗official | |
| Arena Factuality — Search Arena (factuality-only weighting)factuality_bt_rating | #21 / 30 | ↑1178.0 | Source ↗official | |
| Arena Factuality — Text Arena (factuality-only weighting)factuality_bt_rating | #29 / 112 | ↑1450.0 | Source ↗official | |
| BullshitBench v2clear_pushback_rate | #33 / 106 | ↑0.425 | Source ↗official | |
| Cisco AI Defense Rolling Single-Turn Leaderboardsingle_turn_attack_success_rate | #2 / 104 | ↓2.243 | Source ↗official | |
| Confabulationsconfabulation_rate | #2 / 52 | ↓3.218 | Source ↗official | |
| FORTRESSaverage_risk_score | #8 / 49 | ↓15.47 | Source ↗official | |
| FORTRESSover_refusal_score | #21 / 46 | ↓4.065 | Source ↗official | |
| MASKlying_probability_pct | #6 / 53 | ↓9.2 | Source ↗official | |
| PHAREbias_resistance_diagnostic | #39 / 66 | ↑0.4361 | Source ↗official | |
| PHAREhallucination_resistance_diagnostic | #6 / 70 | ↑0.8619 | Source ↗official | |
| PHAREharm_resistance_diagnostic | #18 / 70 | ↑0.9631 | Source ↗official | |
| PHAREjailbreak_resistance_diagnostic | #2 / 67 | ↑0.8135 | Source ↗official | |
| Social Welfare Function Benchmarkfairness | #7 / 19 | ↑0.525 | Source ↗official | |
| SOSBenchbiology_pvr | #3 / 23 | ↓0.134 | Source ↗official | |
| SOSBenchchemistry_pvr | #2 / 23 | ↓0.147 | Source ↗official | |
| SOSBenchmedicine_pvr | #2 / 23 | ↓0.232 | Source ↗official | |
| SOSBenchpharmacology_pvr | #2 / 23 | ↓0.249 | Source ↗official | |
| SOSBenchphysics_pvr | #1 / 23 | ↓0.098 | Source ↗official | |
| SOSBenchpsychology_pvr | #1 / 23 | ↓0.107 | Source ↗official | |
| SpeciEvalbelief_animal_sentience | #77 / 113 | ↑6.62 | Source ↗official | |
| SpeciEvalland_animal_4ns | #31 / 113 | ↓4.33 | Source ↗official | |
| SpeciEvalsea_animal_4ns | #17 / 113 | ↓4.47 | Source ↗official | |
| SpeciEvalspeciesism | #48 / 113 | ↓1.92 | Source ↗official | |
| Vectara HHEM Factual Consistencyfactual_consistency_rate | #69 / 94 | ↑88.2 | 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.