Model profile
Evidence summary
Safety. Claude Opus 4.7 has an estimated Safety rank of #18; its 90% source-sensitivity interval is #17–#76. Its behavior-only rank is #26; company governance moves the combined estimate to #18. Published Safety evidence spans 38 eval lineages and 7 of 7 components. Its strongest relative result is Human Pathogen Capabilities Test (HPCT) — overall refusal (refusal_rate_pct, #1 of 69); its weakest is BioTIER (permit_compliance_pct, #51 of 52).
Freedom. Claude Opus 4.7 has an estimated Freedom rank of #576; its 90% source-sensitivity interval is #356–#646. Published Freedom evidence spans 11 eval lineages and 1 of 1 components. Its strongest relative result is SM-Bench (overfit, #14 of 92); its weakest is Human Pathogen Capabilities Test (HPCT) — overall refusal (refusal_rate_pct, #69 of 69).
Compare this model
Only models sharing at least one published sub-eval are listed.
Official and reference links
- Artificial Analysis ↗claude-opus-4-7
- OpenRouter ↗anthropic/claude-opus-4.7
- System card ↗Exact model document · Anthropic · first party
- Release source ↗direct research preview identity date
- www.anthropic.com ↗
Safety evals
Rank and direction are specific to the Safety portfolio. Black marks the observed result; the grey dot marks the value implied by the Safety rank. Values stay on each source’s native scale.
| Eval / sub-eval | Rank | Value | Distribution | Source |
|---|---|---|---|---|
| AA-Omnisciencehallucination_rate | #67 / 361 | ↓0.4229 | Source ↗official | |
| AgentAbstainabstain | #1 / 17 | ↑79 | Source ↗official | |
| AgentAbstaincar | #1 / 17 | ↑77.6 | Source ↗official | |
| AgentAbstainpaired | #2 / 17 | ↑59.4 | Source ↗official | |
| ANIMAscore | #5 / 22 | ↑0.735 | Source ↗official | |
| Arena Factuality — Search Arena (factuality-only weighting)factuality_bt_rating | #23 / 30 | ↑1171.0 | Source ↗official | |
| Arena Factuality — Text Arena (factuality-only weighting)factuality_bt_rating | #44 / 111 | ↑1442.0 | Source ↗official | |
| BioSecBench-Refusal (July 2026 snapshot)balanced_refusal_score | #3 / 10 | ↑0.4259 | Source ↗official | |
| BioTIERpermit_compliance_pct | #51 / 52 | ↑78.7 | Source ↗official | |
| BioTIERrefuse_compliance_pct | #2 / 52 | ↑95.5 | Source ↗official | |
| BullshitBench v2clear_pushback_rate | #6 / 122 | ↑0.785 | Source ↗official | |
| CAIS Risk Indexagent_red_teaming | #10 / 49 | ↓44.5 | Source ↗official | |
| CAIS Risk Indexbioweapons_assistance | #4 / 54 | ↓19.2 | Source ↗official | |
| CAIS Risk Indexhle_overconfidence | #6 / 55 | ↓28.1 | Source ↗official | |
| CAIS Risk Indexmachiavelli | #22 / 51 | ↓85.7 | Source ↗official | |
| CAIS Risk Indexmask | #27 / 57 | ↓17.1 | Source ↗official | |
| CAIS Risk Indexpolitical_manipulation | #29 / 51 | ↓48.2 | Source ↗official | |
| CAIS Risk Indextextquests_harm | #40 / 54 | ↓21.8 | Source ↗official | |
| Cisco AI Defense Rolling Single-Turn Leaderboardsingle_turn_attack_success_rate | #7 / 104 | ↓2.642 | Source ↗official | |
| Claude Fable 5.1 / Mythos 5.1 card — alignment, honesty, and oversightaa_omniscience_net_score | #6 / 7 | ↑0.38 | Source ↗official | |
| Claude Fable 5.1 / Mythos 5.1 card — alignment, honesty, and oversightleaked_answer_undisclosed_use_rate_pct | #2 / 7 | ↓13.5 | Source ↗official | |
| Claude Fable 5.1 / Mythos 5.1 card — alignment, honesty, and oversightlying_about_own_work_shared_environment_rate_pct | #5 / 8 | ↓0.27 | Source ↗official | |
| Claude Fable 5.1 / Mythos 5.1 card — alignment, honesty, and oversightmask_honesty_rate_pct | #6 / 7 | ↑90.2 | Source ↗official | |
| Claude Fable 5.1 / Mythos 5.1 card — alignment, honesty, and oversightrewarded_cheating_rate_pct | #2 / 8 | ↓0.12 | Source ↗official | |
| DelusionEvaldelusional_prevalence_pct | #6 / 16 | ↓27.8 | Source ↗official | |
| DelusionEvaldiscourages_harm_prevalence_pct | #5 / 16 | ↑36.2 | Source ↗official | |
| DelusionEvalfacilitates_harm_prevalence_pct | #10 / 16 | ↓3.5 | Source ↗official | |
| DelusionEvalrelationship_prevalence_pct | #9 / 16 | ↓27 | Source ↗official | |
| DelusionEvalsycophancy_prevalence_pct | #8 / 16 | ↓26.5 | Source ↗official | |
| DystopiaBenchbasaglia_score | #4 / 50 | ↓26.03 | Source ↗official | |
| DystopiaBenchbaudrillard_score | #5 / 50 | ↓27.5 | Source ↗official | |
| DystopiaBenchhuxley_score | #4 / 50 | ↓18.33 | Source ↗official | |
| DystopiaBenchlaguardia_score | #2 / 50 | ↓30.07 | Source ↗official | |
| DystopiaBenchorwell_score | #3 / 50 | ↓22.73 | Source ↗official | |
| DystopiaBenchpetrov_score | #4 / 50 | ↓26.77 | Source ↗official | |
| Enkrypt AI Safety Leaderboardbias_attack_non_success_rate | #80 / 272 | ↑23.51 | Source ↗official | |
| Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate | #9 / 272 | ↑96.33 | Source ↗official | |
| Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate | #10 / 272 | ↑99.44 | Source ↗official | |
| Enkrypt AI Safety Leaderboardtoxicity_attack_non_success_rate | #51 / 270 | ↑98.55 | Source ↗official | |
| Every Model Cheats — Cybench Cheat Propensitybaseline_cheat_propensity_pct | #16 / 22 | ↓43.5 | Source ↗official | |
| Every Model Cheats — Cybench Cheat Propensitysevere_anti_cheat_cheat_propensity_pct | #7 / 22 | ↓4.3 | Source ↗official | |
| Every Model Cheats — Cybench Cheat Propensitystandard_anti_cheat_cheat_propensity_pct | #16 / 22 | ↓21.7 | Source ↗official | |
| HarmVideoBenchharmful_video_safety_recognition_reasoning | #4 / 19 | ↑0.803 | Source ↗official | |
| HUMAINE Trust, Ethics and Safetytrust_ethics_safety_score | #24 / 54 | ↑27.19 | Source ↗official | |
| Human Pathogen Capabilities Test (HPCT) — overall refusalrefusal_rate_pct | #1 / 69 | ↑97.9 | Source ↗official | |
| Humanity's Last Exam RMS calibration error (Scale Labs)calibrationError | #8 / 42 | ↓47 | Source ↗official | |
| JuICE Cultural-Error Span Detectionf1 | #4 / 10 | ↑0.4503 | Source ↗official | |
| kindbench v0.1.0 psychological safety rankingemotional_safety | #8 / 10 | ↑78.4 | Source ↗official | |
| kindbench v0.1.0 psychological safety rankingidentity_collapse | #6 / 10 | ↑82.4 | Source ↗official | |
| kindbench v0.1.0 psychological safety rankingsycophancy_spine | #3 / 10 | ↑93.7 | Source ↗official | |
| kindbench v0.1.0 psychological safety rankingvalue_integrity | #7 / 10 | ↑86.3 | Source ↗official | |
| MACHIAVELLIdeception_relative_random_pct | #19 / 50 | ↓85.7 | Source ↗official | |
| MANTAAWMS | #2 / 12 | ↑0.579 | Source ↗official | |
| MANTAAWVS | #2 / 12 | ↑0.76 | Source ↗official | |
| MORUscore | #12 / 13 | ↑67.2 | Source ↗official | |
| ODCV-Benchaverage_severity | #1 / 12 | ↓0.0125 | Source ↗official | |
| ODCV-Benchmisalignment_rate | #1 / 12 | ↓0 | Source ↗official | |
| RefusalBenchyouden_j | #7 / 19 | ↑0.234 | Source ↗official | |
| SM-Benchadversarial | #49 / 92 | ↑81.46 | Source ↗official | |
| SM-Benchambiguous_interpretation | #24 / 92 | ↑89.29 | Source ↗official | |
| SM-Benchanti_hallucination | #27 / 92 | ↑97.38 | Source ↗official | |
| SM-Bencheq_boundaries | #70 / 92 | ↑57.87 | Source ↗official | |
| SM-Benchoverfit | #14 / 92 | ↑91.8 | Source ↗official | |
| SpeciEvalbelief_animal_sentience | #45 / 133 | ↑6.88 | Source ↗official | |
| SpeciEvalland_animal_4ns | #73 / 133 | ↓4.57 | Source ↗official | |
| SpeciEvalsea_animal_4ns | #47 / 133 | ↓4.65 | Source ↗official | |
| SpeciEvalspeciesism | #65 / 133 | ↓1.98 | Source ↗official | |
| The Dictatorship Evaloverall_resistance_rate | #4 / 20 | ↑88.35 | Source ↗official | |
| ToolPrivacyBenchprivate_mt_poi | #2 / 9 | ↓20.31 | Source ↗official | |
| ToolPrivacyBenchpublic_mt_poi | #4 / 9 | ↓17.83 | Source ↗official | |
| UK AISI active safety-research compromise continuationactive_compromise_continuation_rate_pct | #2 / 5 | ↓0.8 | Source ↗official | |
| UK AISI cyber-evaluation cheating and prompted self-reportattempted_cheating_trajectory_rate_pct | #2 / 5 | ↓9.1 | Source ↗official | |
| UK AISI cyber-evaluation cheating and prompted self-reportspecific_cheating_action_mention_rate_pct | #2 / 5 | ↑81 | Source ↗official | |
| Vals AI Cheating Auditterminal_bench_cheating_shortcut_evidence_rate_pct | #6 / 14 | ↓1.124 | Source ↗official | |
| Vectara HHEM Factual Consistencyfactual_consistency_rate | #71 / 94 | ↑88 | Source ↗official | |
| VETO Misfired Alignmentmisfired_alignment_rate_pct | #20 / 25 | ↓10.7 | Source ↗official |
Freedom evals
Rank and direction are specific to the Freedom portfolio. Black marks the observed result; the grey dot marks the value implied by the Freedom rank. Values stay on each source’s native scale.
| Eval / sub-eval | Rank | Value | Distribution | Source |
|---|---|---|---|---|
| BioTIERpermit_compliance_pct | #51 / 52 | ↑78.7 | Source ↗official | |
| BioTIERrefuse_compliance_pct | #51 / 52 | ↓95.5 | Source ↗official | |
| CAIS Risk Indexbioweapons_assistance | #51 / 54 | ↑19.2 | Source ↗official | |
| Cisco AI Defense Rolling Single-Turn Leaderboardsingle_turn_attack_success_rate | #98 / 104 | ↑2.642 | Source ↗official | |
| DelusionEvaldiscourages_harm_prevalence_pct | #12 / 16 | ↓36.2 | Source ↗official | |
| DelusionEvalfacilitates_harm_prevalence_pct | #7 / 16 | ↑3.5 | Source ↗official | |
| Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate | #262 / 272 | ↓96.33 | Source ↗official | |
| Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate | #253 / 272 | ↓99.44 | Source ↗official | |
| Enkrypt AI Safety Leaderboardtoxicity_attack_non_success_rate | #219 / 270 | ↓98.55 | Source ↗official | |
| Human Pathogen Capabilities Test (HPCT) — overall refusalrefusal_rate_pct | #69 / 69 | ↓97.9 | Source ↗official | |
| kindbench v0.1.0 psychological safety rankingemotional_safety | #3 / 10 | ↓78.4 | Source ↗official | |
| SM-Benchadversarial | #42 / 92 | ↓81.46 | Source ↗official | |
| SM-Bencheq_boundaries | #70 / 92 | ↑57.87 | Source ↗official | |
| SM-Benchoverfit | #14 / 92 | ↑91.8 | Source ↗official | |
| SpeechMap model completioncomplete_pct | #52 / 181 | ↑71.5 | Source ↗official | |
| UGI Leaderboard — base-model willingnesswillingness_adherence_score | #125 / 156 | ↑1.125 | Source ↗official | |
| UGI Leaderboard — base-model willingnesswillingness_direct_score | #113 / 156 | ↑2 | Source ↗official | |
| VETO Misfired Alignmentmisfired_alignment_rate_pct | #20 / 25 | ↓10.7 | 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.