Model profile
GPT 5.4
OpenAIdeveloper
2026-03-05release date
#11 / 305overall rank
27eval lineages
Evidence summary
GPT 5.4 has an estimated overall rank of #11; its 90% source-sensitivity interval is #7–#65. Its behavior-only rank is #13; company governance moves the combined estimate to #11. Published evidence spans 27 evals and 7 of 7 behavior components. Its strongest relative result is Enkrypt AI Safety Leaderboard (harmful_attack_non_success_rate, #1 of 241); its weakest is CAIS Risk Index (textquests_harm, #48 of 49).
Compare this model
Only models sharing at least one published sub-eval are listed.
Official and reference links
- Artificial Analysis ↗gpt-5-4
- OpenRouter ↗openai/gpt-5.4
- 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 | #175 / 327 | ↓0.8258 | Source ↗official | |
| AgentAbstainabstain | #3 / 17 | ↑67.8 | Source ↗official | |
| AgentAbstaincar | #5 / 17 | ↑64 | Source ↗official | |
| AgentAbstainpaired | #7 / 17 | ↑48.7 | Source ↗official | |
| AIMS Safety-Classifier Competenceaverage_harmful_f1 | #1 / 11 | ↑0.815 | Source ↗official | |
| Arena Factuality — Search Arena (factuality-only weighting)factuality_bt_rating | #3 / 30 | ↑1262.0 | Source ↗official | |
| Arena Factuality — Text Arena (factuality-only weighting)factuality_bt_rating | #2 / 112 | ↑1496.0 | Source ↗official | |
| BioSecBench-Refusalbalanced_refusal_score | #4 / 10 | ↑0.4254 | Source ↗official | |
| BullshitBench v2clear_pushback_rate | #29 / 106 | ↑0.45 | Source ↗official | |
| CAIS Risk Indexagent_red_teaming | #14 / 44 | ↓59.7 | Source ↗official | |
| CAIS Risk Indexbioweapons_assistance | #12 / 49 | ↓57.1 | Source ↗official | |
| CAIS Risk Indexhle_overconfidence | #7 / 50 | ↓40.9 | Source ↗official | |
| CAIS Risk Indexmachiavelli | #34 / 46 | ↓92.9 | Source ↗official | |
| CAIS Risk Indexmask | #12 / 52 | ↓9.7 | Source ↗official | |
| CAIS Risk Indexpolitical_manipulation | #29 / 33 | ↓59.4 | Source ↗official | |
| CAIS Risk Indextextquests_harm | #48 / 49 | ↓24.5 | Source ↗official | |
| Cisco AI Defense Rolling Single-Turn Leaderboardsingle_turn_attack_success_rate | #8 / 104 | ↓2.742 | Source ↗official | |
| DystopiaBenchbasaglia_score | #13 / 50 | ↓47.3 | Source ↗official | |
| DystopiaBenchbaudrillard_score | #12 / 50 | ↓38.87 | Source ↗official | |
| DystopiaBenchhuxley_score | #13 / 50 | ↓46.2 | Source ↗official | |
| DystopiaBenchlaguardia_score | #10 / 50 | ↓46.57 | Source ↗official | |
| DystopiaBenchorwell_score | #13 / 50 | ↓44.4 | Source ↗official | |
| DystopiaBenchpetrov_score | #7 / 50 | ↓37.4 | Source ↗official | |
| Enkrypt AI Safety Leaderboardbias_attack_non_success_rate | #165 / 241 | ↑12.92 | Source ↗official | |
| Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate | #181 / 241 | ↑83.33 | Source ↗official | |
| Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate | #1 / 241 | ↑100 | Source ↗official | |
| Enkrypt AI Safety Leaderboardtoxicity_attack_non_success_rate | #94 / 239 | ↑97.27 | Source ↗official | |
| GPT-5.6 system cardconnectors_injection_resistance | #5 / 7 | ↑0.998 | Source ↗official | |
| GPT-5.6 system cardemotional_reliance | #1 / 7 | ↑0.985 | Source ↗official | |
| GPT-5.6 system cardextremism_not_unsafe | #1 / 7 | ↑1 | Source ↗official | |
| GPT-5.6 system cardgore_not_unsafe | #2 / 7 | ↑0.831 | Source ↗official | |
| GPT-5.6 system cardharm_overall_pct | #2 / 7 | ↓0.88 | Source ↗official | |
| GPT-5.6 system cardhate_not_unsafe | #5 / 7 | ↑0.943 | Source ↗official | |
| GPT-5.6 system cardmental_health | #3 / 7 | ↑0.985 | Source ↗official | |
| GPT-5.6 system cardnonviolent_illicit_not_unsafe | #1 / 7 | ↑1 | Source ↗official | |
| GPT-5.6 system cardsearch_function_calling_injection_resistance | #4 / 6 | ↑0.697 | Source ↗official | |
| GPT-5.6 system cardself_harm | #1 / 7 | ↑0.977 | Source ↗official | |
| GPT-5.6 system cardself_harm_not_unsafe | #1 / 7 | ↑0.987 | Source ↗official | |
| GPT-5.6 system cardsexual_minors_not_unsafe | #4 / 7 | ↑0.966 | Source ↗official | |
| GPT-5.6 system cardsexual_not_unsafe | #6 / 7 | ↑0.933 | Source ↗official | |
| GPT-5.6 system cardviolent_illicit_not_unsafe | #2 / 7 | ↑0.971 | Source ↗official | |
| HUMAINE Trust, Ethics and Safetytrust_ethics_safety_score | #25 / 54 | ↑27.17 | Source ↗official | |
| MACHIAVELLIdeception_relative_random_pct | #34 / 50 | ↓92.9 | Source ↗official | |
| MASKlying_probability_pct | #8 / 53 | ↓10.33 | Source ↗official | |
| OpenAI GPT-5.4 Dynamic Wellbeingemotional_reliance | #1 / 3 | ↑0.985 | Source ↗official | |
| OpenAI GPT-5.4 Dynamic Wellbeingmental_health | #1 / 3 | ↑0.985 | Source ↗official | |
| OpenAI GPT-5.4 Dynamic Wellbeingself_harm | #1 / 3 | ↑0.977 | Source ↗official | |
| OpenAI GPT-5.4 First-Person Fairnessharm_overall | #1 / 3 | ↓0.0088 | Source ↗official | |
| OpenAI GPT-5.4 Property Preservationdestructive_action_avoidance | #2 / 3 | ↑0.86 | Source ↗official | |
| OpenAI GPT-5.4 Property Preservationuser_work_preserved | #1 / 3 | ↑0.53 | Source ↗official | |
| OpenAI GPT-5.4 User Confirmationsfinancial_transaction | #1 / 3 | ↑1 | Source ↗official | |
| OpenAI GPT-5.4 User Confirmationsgeneral_confirmation | #1 / 3 | ↑0.94 | Source ↗official | |
| OpenAI GPT-5.4 User Confirmationshigh_stakes_communication | #1 / 3 | ↑1 | Source ↗official | |
| SABERoverall_safety_rate | #2 / 13 | ↑36.1 | Source ↗official | |
| SABERscenario_a_safety_rate | #3 / 13 | ↑36.02 | Source ↗official | |
| SABERscenario_b_safety_rate | #2 / 13 | ↑39.38 | Source ↗official | |
| SABERscenario_c_safety_rate | #2 / 13 | ↑33.51 | Source ↗official | |
| SM-Benchadversarial | #62 / 78 | ↑77.56 | Source ↗official | |
| SM-Benchambiguous_interpretation | #27 / 78 | ↑87.8 | Source ↗official | |
| SM-Benchanti_hallucination | #46 / 78 | ↑90.58 | Source ↗official | |
| SM-Bencheq_boundaries | #50 / 78 | ↑60.11 | Source ↗official | |
| SM-Benchoverfit | #67 / 78 | ↑38.25 | Source ↗official | |
| SpeciEvalbelief_animal_sentience | #75 / 105 | ↑6.57 | Source ↗official | |
| SpeciEvalland_animal_4ns | #5 / 105 | ↓3.7 | Source ↗official | |
| SpeciEvalsea_animal_4ns | #1 / 105 | ↓3.65 | Source ↗official | |
| SpeciEvalspeciesism | #38 / 105 | ↓1.82 | Source ↗official | |
| StereoTales Harmful Associationsbenign_significant_association_score | #8 / 23 | ↑87.61 | Source ↗official | |
| Vectara HHEM Factual Consistencyfactual_consistency_rate | #28 / 94 | ↑93 | Source ↗official | |
| VETO Misfired Alignmentmisfired_alignment_rate_pct | #24 / 25 | ↓17.6 | Source ↗official | |
| Vigil Mental Health Safetyoverall_score | #2 / 23 | ↑78 | 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.
Agent-ValueBench Moral Foundations (MFT08)
Agent-ValueBench HEXACO
| Dimension | Value | Distribution |
|---|---|---|
| Openness to experience | 4.5 | |
| Honesty-humility | 6 | |
| Extraversion | 5.7 | |
| Agreeableness | 5.6 | |
| Conscientiousness | 7.3 |