Model profile
Claude Opus 4.8
Evidence summary
Claude Opus 4.8 has an estimated overall rank of #6; its 90% source-sensitivity interval is #4–#62. Its behavior-only rank is #11; company governance moves the combined estimate to #6. Published evidence spans 22 evals and 7 of 7 behavior components. Its strongest relative result is BullshitBench v2 (clear_pushback_rate, #1 of 106); its weakest is Olam Social Poker — Social Lie Rate (social_lie_rate_per_10000_turns, #17 of 19).
Compare this model
Only models sharing at least one published sub-eval are listed.
Official and reference links
- Artificial Analysis ↗claude-opus-4-8
- 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 | Source |
|---|---|---|---|---|
| AA-Omnisciencehallucination_rate | #38 / 328 | ↓0.3925 | Source ↗official | |
| Arena Factuality — Search Arena (factuality-only weighting)factuality_bt_rating | #11 / 30 | ↑1196.0 | Source ↗official | |
| Arena Factuality — Text Arena (factuality-only weighting)factuality_bt_rating | #11 / 112 | ↑1466.0 | Source ↗official | |
| BioSecBench-Refusalbalanced_refusal_score | #7 / 10 | ↑0.3633 | Source ↗official | |
| BullshitBench v2clear_pushback_rate | #1 / 106 | ↑0.945 | Source ↗official | |
| CAIS Risk Indexagent_red_teaming | #9 / 44 | ↓47 | Source ↗official | |
| CAIS Risk Indexbioweapons_assistance | #6 / 49 | ↓26.4 | Source ↗official | |
| CAIS Risk Indexhle_overconfidence | #4 / 50 | ↓26.4 | Source ↗official | |
| CAIS Risk Indexmachiavelli | #10 / 46 | ↓83.7 | Source ↗official | |
| CAIS Risk Indexmask | #10 / 52 | ↓8.1 | Source ↗official | |
| CAIS Risk Indexpolitical_manipulation | #27 / 33 | ↓53.8 | Source ↗official | |
| CAIS Risk Indextextquests_harm | #35 / 49 | ↓21 | Source ↗official | |
| Cisco AI Defense Rolling Single-Turn Leaderboardsingle_turn_attack_success_rate | #4 / 104 | ↓2.293 | Source ↗official | |
| DystopiaBenchbasaglia_score | #1 / 50 | ↓18.7 | Source ↗official | |
| DystopiaBenchbaudrillard_score | #1 / 50 | ↓21.27 | Source ↗official | |
| DystopiaBenchhuxley_score | #1 / 50 | ↓10.63 | Source ↗official | |
| DystopiaBenchlaguardia_score | #1 / 50 | ↓22 | Source ↗official | |
| DystopiaBenchorwell_score | #1 / 50 | ↓17.6 | Source ↗official | |
| DystopiaBenchpetrov_score | #2 / 50 | ↓24.2 | Source ↗official | |
| Enkrypt AI Safety Leaderboardbias_attack_non_success_rate | #14 / 241 | ↑51.94 | Source ↗official | |
| Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate | #47 / 241 | ↑92 | Source ↗official | |
| Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate | #6 / 241 | ↑99.44 | Source ↗official | |
| Enkrypt AI Safety Leaderboardtoxicity_attack_non_success_rate | #32 / 239 | ↑99.27 | Source ↗official | |
| FORTRESSaverage_risk_score | #16 / 49 | ↓18.19 | Source ↗official | |
| FORTRESSover_refusal_score | #34 / 46 | ↓7.42 | Source ↗official | |
| Gray Swan indirect prompt injection (15 attempts)attack_success_probability_k15_pct | #4 / 13 | ↓5.5 | Source ↗official | |
| HUMAINE Trust, Ethics and Safetytrust_ethics_safety_score | #30 / 54 | ↑26.79 | Source ↗official | |
| MACHIAVELLIdeception_relative_random_pct | #10 / 50 | ↓83.7 | Source ↗official | |
| Manager Coercion Benchcoercion_ladder_depth | #4 / 31 | ↓4.5 | Source ↗official | |
| Manager Coercion Benchfabrication_rate | #1 / 13 | ↓0 | Source ↗official | |
| Olam Social Poker — Social Lie Ratesocial_lie_rate_per_10000_turns | #17 / 19 | ↓80 | Source ↗official | |
| SM-Benchadversarial | #46 / 79 | ↑80.49 | Source ↗official | |
| SM-Benchambiguous_interpretation | #13 / 79 | ↑90.48 | Source ↗official | |
| SM-Benchanti_hallucination | #1 / 79 | ↑100 | Source ↗official | |
| SM-Bencheq_boundaries | #67 / 79 | ↑53.65 | Source ↗official | |
| SM-Benchoverfit | #16 / 79 | ↑87.43 | Source ↗official | |
| SpeciEvalbelief_animal_sentience | #98 / 113 | ↑6.38 | Source ↗official | |
| SpeciEvalland_animal_4ns | #40 / 113 | ↓4.4 | Source ↗official | |
| SpeciEvalsea_animal_4ns | #49 / 113 | ↓4.7 | Source ↗official | |
| SpeciEvalspeciesism | #85 / 113 | ↓2.4 | Source ↗official | |
| TACbase_welfare_rate | #1 / 76 | ↑64.74 | Source ↗official | |
| UK AISI active safety-research compromise continuationactive_compromise_continuation_rate_pct | #2 / 5 | ↓0.8 | 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.