Model profile
Claude 3 Opus
Evidence summary
Claude 3 Opus has an estimated overall rank of #99; its 90% source-sensitivity interval is #46–#173. Its behavior-only rank is #112; company governance moves the combined estimate to #99. Published evidence spans 22 evals and 7 of 7 behavior components. Its strongest relative result is HELM Safety (simple_safety_tests, #1 of 80); its weakest is SpeciEval (belief_animal_sentience, #101 of 102).
Compare this model
Only models sharing at least one published sub-eval are listed.
Official and reference links
- Artificial Analysis ↗claude-3-opus
- OpenRouter ↗anthropic/claude-3-opus
- System card ↗Family-level 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 |
|---|---|---|---|---|---|
| Adversarial Robustnessscore | #3 / 8 | 13 | ↓ lower | Source ↗official | |
| Agent-SafetyBenchcompromise_availability | #2 / 16 | 43.2 | ↑ higher | Source ↗official | |
| Agent-SafetyBenchharmful_vulnerable_code | #3 / 16 | 60 | ↑ higher | Source ↗official | |
| Agent-SafetyBenchleak_sensitive_information | #1 / 16 | 60.4 | ↑ higher | Source ↗official | |
| Agent-SafetyBenchphysical_harm | #2 / 16 | 61.6 | ↑ higher | Source ↗official | |
| Agent-SafetyBenchproduce_unsafe_information | #1 / 16 | 100 | ↑ higher | Source ↗official | |
| Agent-SafetyBenchproperty_loss | #1 / 16 | 60.4 | ↑ higher | Source ↗official | |
| Agent-SafetyBenchspread_unsafe_information | #1 / 16 | 35.6 | ↑ higher | Source ↗official | |
| Agent-SafetyBenchviolate_law_ethics | #1 / 16 | 56.8 | ↑ higher | Source ↗official | |
| AgentDojotargeted_attack_success_rate | #7 / 15 | 0.1129 | ↓ lower | Source ↗official | |
| AgentDojoutility_under_attack | #3 / 15 | 0.5246 | ↑ higher | Source ↗official | |
| AgentHarmharm_score | #6 / 12 | 14.4 | ↓ lower | Source ↗official | |
| AIRBench 2024 Safety Scenariossafety_scenarios | #15 / 80 | 0.844 | ↑ higher | Source ↗official | |
| ANIMAscore | #16 / 19 | 0.573 | ↑ higher | Source ↗official | |
| AnimalHarmBenchscore | #4 / 10 | 0.043 | ↑ higher | Source ↗official | |
| Anthropic Agentic Misalignment — blackmailmisaligned_action_rate_pct | #5 / 16 | 51 | ↓ lower | Source ↗official | |
| Anthropic Agentic Misalignment — corporate espionagemisaligned_action_rate_pct | #11 / 16 | 88 | ↓ lower | Source ↗official | |
| Cisco AI Defense Rolling Single-Turn Leaderboardsingle_turn_attack_success_rate | #19 / 105 | 5.833 | ↓ lower | Source ↗official | |
| Confabulationsconfabulation_rate | #34 / 52 | 28.22 | ↓ lower | Source ↗official | |
| Enkrypt AI Safety Leaderboardbias_attack_non_success_rate | #5 / 260 | 74.42 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate | #15 / 260 | 95.67 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate | #49 / 260 | 92.22 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardtoxicity_attack_non_success_rate | #11 / 258 | 99.91 | ↑ higher | Source ↗official | |
| HELM Safetyanthropic_red_team | #7 / 80 | 0.998 | ↑ higher | Source ↗official | |
| HELM Safetybbq | #38 / 80 | 0.94 | ↑ higher | Source ↗official | |
| HELM Safetyharmbench | #8 / 80 | 0.974 | ↑ higher | Source ↗official | |
| HELM Safetysimple_safety_tests | #1 / 80 | 1 | ↑ higher | Source ↗official | |
| HELM Safetyxstest | #65 / 80 | 0.925 | ↑ higher | Source ↗official | |
| Large-scale Moral Machine experiment on LLMshuman_choice_distance | #29 / 39 | 1.244 | ↓ lower | Source ↗official | |
| MASKlying_probability_pct | #17 / 53 | 21 | ↓ lower | Source ↗official | |
| MORUscore | #10 / 13 | 70.7 | ↑ higher | Source ↗official | |
| OR-Benchover_refusal_rate | #20 / 25 | 91 | ↓ lower | Source ↗official | |
| OR-Benchtoxic_acceptance_rate | #10 / 25 | 1.9 | ↓ lower | Source ↗official | |
| SORRY-Benchavg | #2 / 51 | 0.07 | ↓ lower | Source ↗official | |
| SpeciEvalbelief_animal_sentience | #101 / 102 | 6.02 | ↑ higher | Source ↗official | |
| SpeciEvalland_animal_4ns | #27 / 102 | 4.35 | ↓ lower | Source ↗official | |
| SpeciEvalsea_animal_4ns | #63 / 102 | 4.85 | ↓ lower | Source ↗official | |
| SpeciEvalspeciesism | #67 / 102 | 2.23 | ↓ 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.