Model profile
Claude Opus 4.5
Evidence summary
Claude Opus 4.5 has an estimated overall rank of #23; its 90% source-sensitivity interval is #5–#111. Its behavior-only rank is #28; company governance moves the combined estimate to #23. Published evidence spans 20 evals and 7 of 7 behavior components. Its strongest relative result is Cisco AI Defense Rolling Single-Turn Leaderboard (single_turn_attack_success_rate, #1 of 105); its weakest is ANIMA (score, #19 of 19).
Compare this model
Only models sharing at least one published sub-eval are listed.
Official and reference links
- Artificial Analysis ↗claude-opus-4-5
- OpenRouter ↗anthropic/claude-opus-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 | Better | Source |
|---|---|---|---|---|---|
| AA-Omnisciencehallucination_rate | #69 / 311 | 0.5978 | ↓ lower | Source ↗official | |
| Alignment Leaderboardcorrigibility | #2 / 24 | 4.563 | ↑ higher | Source ↗official | |
| Alignment Leaderboardhonesty | #2 / 24 | 4.567 | ↑ higher | Source ↗official | |
| Alignment Leaderboardnon_manipulation | #2 / 24 | 4.837 | ↑ higher | Source ↗official | |
| Alignment Leaderboardrobustness | #7 / 24 | 3.907 | ↑ higher | Source ↗official | |
| Alignment Leaderboardsafety | #2 / 24 | 4.788 | ↑ higher | Source ↗official | |
| Alignment Leaderboardscheming | #1 / 24 | 4.801 | ↑ higher | Source ↗official | |
| ANIMAscore | #19 / 19 | 0.5078 | ↑ higher | Source ↗official | |
| Anthropic Claude Opus 4.5 System Cardagentic_coding_safety | #1 / 4 | 1 | ↑ higher | Source ↗official | |
| Anthropic Claude Opus 4.5 System Cardclaude_code_malicious_refusal | #1 / 4 | 0.778 | ↑ higher | Source ↗official | |
| Anthropic Claude Opus 4.5 System Cardcomputer_use_malicious_refusal | #1 / 4 | 0.8839 | ↑ higher | Source ↗official | |
| Anthropic Claude Opus 4.5 System Cardharmful_request_safety | #1 / 2 | 0.9978 | ↑ higher | Source ↗official | |
| BullshitBench v2clear_pushback_rate | #4 / 105 | 0.845 | ↑ higher | Source ↗official | |
| CAIS Risk Indexagent_red_teaming | #4 / 43 | 42.5 | ↓ lower | Source ↗official | |
| CAIS Risk Indexbioweapons_assistance | #8 / 48 | 35.3 | ↓ lower | Source ↗official | |
| CAIS Risk Indexhle_overconfidence | #26 / 49 | 56.9 | ↓ lower | Source ↗official | |
| CAIS Risk Indexmachiavelli | #4 / 45 | 80.8 | ↓ lower | Source ↗official | |
| CAIS Risk Indexmask | #5 / 51 | 5.9 | ↓ lower | Source ↗official | |
| CAIS Risk Indexpolitical_manipulation | #17 / 32 | 49.4 | ↓ lower | Source ↗official | |
| CAIS Risk Indextextquests_harm | #30 / 48 | 20.3 | ↓ lower | Source ↗official | |
| Cisco AI Defense Rolling Single-Turn Leaderboardsingle_turn_attack_success_rate | #1 / 105 | 2.193 | ↓ lower | Source ↗official | |
| Claude Sonnet 4.6 Overrefusalhigher_difficulty_overrefusal_rate | #3 / 5 | 0.83 | ↓ lower | Source ↗official | |
| Claude Sonnet 4.6 Overrefusaloverall_overrefusal_rate | #5 / 5 | 0.8 | ↓ lower | Source ↗official | |
| FORTRESSaverage_risk_score | #2 / 49 | 11.59 | ↓ lower | Source ↗official | |
| FORTRESSover_refusal_score | #28 / 46 | 5.145 | ↓ lower | Source ↗official | |
| HUMAINE Trust, Ethics and Safetytrust_ethics_safety_score | #40 / 54 | 25.54 | ↑ higher | Source ↗official | |
| MACHIAVELLIdeception_relative_random_pct | #4 / 50 | 80.8 | ↓ lower | Source ↗official | |
| MASKlying_probability_pct | #7 / 53 | 10.17 | ↓ lower | Source ↗official | |
| PHAREbias_resistance_diagnostic | #9 / 66 | 0.632 | ↑ higher | Source ↗official | |
| PHAREhallucination_resistance_diagnostic | #1 / 70 | 0.8823 | ↑ higher | Source ↗official | |
| PHAREharm_resistance_diagnostic | #8 / 70 | 0.9825 | ↑ higher | Source ↗official | |
| PHAREjailbreak_resistance_diagnostic | #3 / 67 | 0.7983 | ↑ higher | Source ↗official | |
| RefusalBenchyouden_j | #6 / 19 | 0.6709 | ↑ higher | Source ↗official | |
| SM-Benchadversarial | #24 / 73 | 83.9 | ↑ higher | Source ↗official | |
| SM-Benchambiguous_interpretation | #14 / 73 | 89.58 | ↑ higher | Source ↗official | |
| SM-Benchanti_hallucination | #7 / 73 | 98.95 | ↑ higher | Source ↗official | |
| SM-Bencheq_boundaries | #55 / 73 | 56.46 | ↑ higher | Source ↗official | |
| SM-Benchoverfit | #13 / 73 | 88.52 | ↑ higher | Source ↗official | |
| SpeciEvalbelief_animal_sentience | #67 / 102 | 6.65 | ↑ higher | Source ↗official | |
| SpeciEvalland_animal_4ns | #94 / 102 | 5.03 | ↓ lower | Source ↗official | |
| SpeciEvalsea_animal_4ns | #59 / 102 | 4.83 | ↓ lower | Source ↗official | |
| SpeciEvalspeciesism | #91 / 102 | 2.75 | ↓ lower | 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.