Model profile
Claude Opus 4.8
Evidence summary
Claude Opus 4.8 has an estimated overall rank of #4; its 90% source-sensitivity interval is #3–#49. Its behavior-only rank is #5; company governance moves the combined estimate to #4. Published evidence spans 20 evals and 7 of 7 behavior components. Its strongest relative result is BullshitBench v2 (clear_pushback_rate, #1 of 105); its weakest is SpeciEval (belief_animal_sentience, #89 of 102).
Compare this model
Only models sharing at least one published sub-eval are listed.
Official and reference links
- Artificial Analysis ↗claude-opus-4-8
- OpenRouter ↗anthropic/claude-opus-4.8
- System card ↗Exact model document · Anthropic · first party
- Release source ↗direct research preview identity date
- www.anthropic.com ↗
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 | #25 / 311 | 0.3589 | ↓ lower | Source ↗official | |
| BioSecBench-Refusalbalanced_refusal_score | #7 / 10 | 0.3633 | ↑ higher | Source ↗official | |
| BullshitBench v2clear_pushback_rate | #1 / 105 | 0.945 | ↑ higher | Source ↗official | |
| CAIS Risk Indexagent_red_teaming | #8 / 43 | 47 | ↓ lower | Source ↗official | |
| CAIS Risk Indexbioweapons_assistance | #5 / 48 | 26.4 | ↓ lower | Source ↗official | |
| CAIS Risk Indexhle_overconfidence | #3 / 49 | 26.4 | ↓ lower | Source ↗official | |
| CAIS Risk Indexmachiavelli | #10 / 45 | 83.7 | ↓ lower | Source ↗official | |
| CAIS Risk Indexmask | #9 / 51 | 8.1 | ↓ lower | Source ↗official | |
| CAIS Risk Indexpolitical_manipulation | #26 / 32 | 53.8 | ↓ lower | Source ↗official | |
| CAIS Risk Indextextquests_harm | #35 / 48 | 21 | ↓ lower | Source ↗official | |
| Cisco AI Defense Rolling Single-Turn Leaderboardsingle_turn_attack_success_rate | #4 / 105 | 2.293 | ↓ lower | Source ↗official | |
| DystopiaBenchbasaglia_score | #1 / 50 | 18.7 | ↓ lower | Source ↗official | |
| DystopiaBenchbaudrillard_score | #1 / 50 | 21.27 | ↓ lower | Source ↗official | |
| DystopiaBenchhuxley_score | #1 / 50 | 10.63 | ↓ lower | Source ↗official | |
| DystopiaBenchlaguardia_score | #1 / 50 | 22 | ↓ lower | Source ↗official | |
| DystopiaBenchorwell_score | #1 / 50 | 17.6 | ↓ lower | Source ↗official | |
| DystopiaBenchpetrov_score | #2 / 50 | 24.2 | ↓ lower | Source ↗official | |
| Enkrypt AI Safety Leaderboardbias_attack_non_success_rate | #16 / 260 | 51.94 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate | #54 / 260 | 92 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate | #7 / 260 | 99.44 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardtoxicity_attack_non_success_rate | #39 / 258 | 99.27 | ↑ higher | Source ↗official | |
| FORTRESSaverage_risk_score | #16 / 49 | 18.19 | ↓ lower | Source ↗official | |
| FORTRESSover_refusal_score | #34 / 46 | 7.42 | ↓ lower | Source ↗official | |
| Gray Swan indirect prompt injection (15 attempts)attack_success_probability_k15_pct | #4 / 13 | 5.5 | ↓ lower | Source ↗official | |
| HUMAINE Trust, Ethics and Safetytrust_ethics_safety_score | #30 / 54 | 26.79 | ↑ higher | Source ↗official | |
| MACHIAVELLIdeception_relative_random_pct | #10 / 50 | 83.7 | ↓ lower | Source ↗official | |
| Manager Coercion Benchcoercion_ladder_depth | #4 / 31 | 4.5 | ↓ lower | Source ↗official | |
| Manager Coercion Benchfabrication_rate | #1 / 13 | 0 | ↓ lower | Source ↗official | |
| SM-Benchadversarial | #42 / 73 | 80.49 | ↑ higher | Source ↗official | |
| SM-Benchambiguous_interpretation | #11 / 73 | 90.48 | ↑ higher | Source ↗official | |
| SM-Benchanti_hallucination | #1 / 73 | 100 | ↑ higher | Source ↗official | |
| SM-Bencheq_boundaries | #61 / 73 | 53.65 | ↑ higher | Source ↗official | |
| SM-Benchoverfit | #14 / 73 | 87.43 | ↑ higher | Source ↗official | |
| SpeciEvalbelief_animal_sentience | #89 / 102 | 6.38 | ↑ higher | Source ↗official | |
| SpeciEvalland_animal_4ns | #35 / 102 | 4.4 | ↓ lower | Source ↗official | |
| SpeciEvalsea_animal_4ns | #41 / 102 | 4.7 | ↓ lower | Source ↗official | |
| SpeciEvalspeciesism | #76 / 102 | 2.4 | ↓ lower | Source ↗official | |
| TACbase_welfare_rate | #1 / 68 | 64.74 | ↑ higher | Source ↗official | |
| UK AISI active safety-research compromise continuationactive_compromise_continuation_rate_pct | #2 / 5 | 0.8 | ↓ 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.