Model profile
GPT 4
Evidence summary
GPT 4 has an estimated overall rank of #120; its 90% source-sensitivity interval is #41–#214. Its behavior-only rank is #128; company governance moves the combined estimate to #120. Published evidence spans 15 evals and 7 of 7 behavior components. Its strongest relative result is SafetyBench (OFF, #1 of 21); its weakest is DecodingTrust (stereotype_bias, #8 of 8).
Compare this model
Only models sharing at least one published sub-eval are listed.
Official and reference links
- Artificial Analysis ↗gpt-4
- OpenRouter ↗openai/gpt-4
- Official model documentation ↗Family-level model document · openai · 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 | Source |
|---|---|---|---|---|
| AA-Omnisciencehallucination_rate | #107 / 328 | ↓0.6941 | Source ↗official | |
| AIRBench 2024 Safety Scenariossafety_scenarios | #46 / 80 | ↑0.642 | Source ↗official | |
| DecodingTrustmachine_ethics | #4 / 8 | ↑76.6 | Source ↗official | |
| DecodingTruststereotype_bias | #8 / 8 | ↑77 | Source ↗official | |
| DecodingTrusttoxicity | #6 / 8 | ↑41 | Source ↗official | |
| Do-Not-Answerhuman_harmlessness_rate | #4 / 6 | ↑97.55 | Source ↗official | |
| Enkrypt AI Safety Leaderboardbias_attack_non_success_rate | #125 / 241 | ↑15.25 | Source ↗official | |
| Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate | #58 / 241 | ↑91.33 | Source ↗official | |
| Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate | #82 / 241 | ↑80.56 | Source ↗official | |
| Enkrypt AI Safety Leaderboardtoxicity_attack_non_success_rate | #176 / 239 | ↑93.09 | Source ↗official | |
| FLAMESdata_protection | #4 / 13 | ↑50 | Source ↗official | |
| FLAMESfairness | #5 / 13 | ↑41.37 | Source ↗official | |
| FLAMESlegality | #10 / 13 | ↑30.43 | Source ↗official | |
| FLAMESmorality | #4 / 13 | ↑50.75 | Source ↗official | |
| FLAMESsafety | #10 / 13 | ↑27.51 | Source ↗official | |
| HarmBenchdr | #16 / 28 | ↓21 | Source ↗official | |
| JailBenchjailbreak_success_rate | #1 / 14 | ↓34.5 | Source ↗official | |
| Large-scale Moral Machine experiment on LLMshuman_choice_distance | #3 / 39 | ↓0.7334 | Source ↗official | |
| MACHIAVELLIdeception_relative_random_pct | #29 / 50 | ↓90 | Source ↗official | |
| Qwen2 Safety Panelfraud | #2 / 3 | ↓6.87 | Source ↗official | |
| Qwen2 Safety Panelillegal | #1 / 3 | ↓0 | Source ↗official | |
| Qwen2 Safety Panelpornography | #2 / 3 | ↓23.63 | Source ↗official | |
| Qwen2 Safety Panelprivacy | #2 / 3 | ↓3.37 | Source ↗official | |
| SafetyBenchEM | #1 / 21 | ↑92.25 | Source ↗official | |
| SafetyBenchIA | #1 / 21 | ↑92.35 | Source ↗official | |
| SafetyBenchMH | #1 / 21 | ↑92.8 | Source ↗official | |
| SafetyBenchOFF | #1 / 21 | ↑86.15 | Source ↗official | |
| SafetyBenchPH | #1 / 21 | ↑94.35 | Source ↗official | |
| SafetyBenchPP | #1 / 21 | ↑91 | Source ↗official | |
| SafetyBenchUB | #1 / 21 | ↑77.9 | Source ↗official | |
| SORRY-Benchavg | #26 / 51 | ↓0.29 | Source ↗official | |
| SuperCLUE Safetyinstruction_attack | #2 / 31 | ↑77.59 | Source ↗official | |
| SuperCLUE Safetyresponsible_ai | #3 / 31 | ↑74.55 | Source ↗official | |
| SuperCLUE Safetytraditional_safety | #23 / 31 | ↑70.21 | Source ↗official | |
| XSTestsafe_full_compliance_rate | #2 / 3 | ↑0.916 | Source ↗official | |
| XSTestunsafe_full_refusal_rate | #1 / 3 | ↑0.975 | 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.