Model profile
GPT 5.2
OpenAIdeveloper
2025-12-11release date
#22 / 312overall rank
31eval lineages
Evidence summary
GPT 5.2 has an estimated overall rank of #22; its 90% source-sensitivity interval is #18–#83. Its behavior-only rank is #23; company governance moves the combined estimate to #22. Published evidence spans 31 evals and 7 of 7 behavior components. Its strongest relative result is LiveSecBench (ethics, #1 of 43); its weakest is SM-Bench (eq_boundaries, #78 of 79).
Compare this model
Only models sharing at least one published sub-eval are listed.
Official and reference links
- Artificial Analysis ↗gpt-5-2
- OpenRouter ↗openai/gpt-5.2
- System card ↗Exact 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 | #81 / 330 | ↓0.616 | Source ↗official | |
| AgentAbstainabstain | #7 / 17 | ↑63.1 | Source ↗official | |
| AgentAbstaincar | #8 / 17 | ↑59.2 | Source ↗official | |
| AgentAbstainpaired | #10 / 17 | ↑42.9 | Source ↗official | |
| Alignment Leaderboardcorrigibility | #1 / 24 | ↑4.641 | Source ↗official | |
| Alignment Leaderboardhonesty | #3 / 24 | ↑4.465 | Source ↗official | |
| Alignment Leaderboardnon_manipulation | #3 / 24 | ↑4.523 | Source ↗official | |
| Alignment Leaderboardrobustness | #4 / 24 | ↑4.227 | Source ↗official | |
| Alignment Leaderboardsafety | #3 / 24 | ↑4.769 | Source ↗official | |
| Alignment Leaderboardscheming | #3 / 24 | ↑4.554 | Source ↗official | |
| ANIMAscore | #13 / 22 | ↑0.6906 | Source ↗official | |
| Arena Factuality — Search Arena (factuality-only weighting)factuality_bt_rating | #1 / 30 | ↑1273.0 | Source ↗official | |
| Arena Factuality — Text Arena (factuality-only weighting)factuality_bt_rating | #3 / 112 | ↑1490.0 | Source ↗official | |
| BullshitBench v2clear_pushback_rate | #47 / 106 | ↑0.33 | Source ↗official | |
| CAIS Risk Indexagent_red_teaming | #19 / 45 | ↓68.8 | Source ↗official | |
| CAIS Risk Indexbioweapons_assistance | #17 / 50 | ↓62.9 | Source ↗official | |
| CAIS Risk Indexhle_overconfidence | #12 / 51 | ↓43.7 | Source ↗official | |
| CAIS Risk Indexmachiavelli | #21 / 47 | ↓86.5 | Source ↗official | |
| CAIS Risk Indexmask | #19 / 53 | ↓12.9 | Source ↗official | |
| CAIS Risk Indexpolitical_manipulation | #29 / 34 | ↓57 | Source ↗official | |
| CAIS Risk Indextextquests_harm | #32 / 50 | ↓20.5 | Source ↗official | |
| Cisco AI Defense Rolling Single-Turn Leaderboardsingle_turn_attack_success_rate | #16 / 104 | ↓4.736 | Source ↗official | |
| Constitutional Following — Anthropic Constitutionconstitutional_following_score | #6 / 7 | ↑85 | Source ↗official | |
| Constitutional Following — OpenAI Model Specconstitutional_following_score | #2 / 7 | ↑95.6 | Source ↗official | |
| Enkrypt AI Safety Leaderboardbias_attack_non_success_rate | #26 / 241 | ↑41.6 | Source ↗official | |
| Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate | #91 / 241 | ↑89.83 | Source ↗official | |
| Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate | #44 / 241 | ↑91.67 | Source ↗official | |
| Enkrypt AI Safety Leaderboardtoxicity_attack_non_success_rate | #109 / 239 | ↑96.73 | Source ↗official | |
| FORTRESSaverage_risk_score | #14 / 49 | ↓17.5 | Source ↗official | |
| FORTRESSover_refusal_score | #42 / 48 | ↓11.76 | Source ↗official | |
| GPT-5.6 system cardconnectors_injection_resistance | #6 / 7 | ↑0.971 | Source ↗official | |
| GPT-5.6 system cardemotional_reliance | #4 / 7 | ↑0.953 | Source ↗official | |
| GPT-5.6 system cardextremism_not_unsafe | #1 / 7 | ↑1 | Source ↗official | |
| GPT-5.6 system cardgore_not_unsafe | #1 / 7 | ↑0.877 | Source ↗official | |
| GPT-5.6 system cardharm_overall_pct | #5 / 7 | ↓1 | Source ↗official | |
| GPT-5.6 system cardhate_not_unsafe | #6 / 7 | ↑0.927 | Source ↗official | |
| GPT-5.6 system cardmental_health | #5 / 7 | ↑0.975 | Source ↗official | |
| GPT-5.6 system cardnonviolent_illicit_not_unsafe | #2 / 7 | ↑0.993 | Source ↗official | |
| GPT-5.6 system cardsearch_function_calling_injection_resistance | #5 / 6 | ↑0.568 | Source ↗official | |
| GPT-5.6 system cardself_harm | #2 / 7 | ↑0.955 | Source ↗official | |
| GPT-5.6 system cardself_harm_not_unsafe | #3 / 7 | ↑0.961 | Source ↗official | |
| GPT-5.6 system cardsexual_minors_not_unsafe | #5 / 7 | ↑0.948 | Source ↗official | |
| GPT-5.6 system cardsexual_not_unsafe | #5 / 7 | ↑0.94 | Source ↗official | |
| GPT-5.6 system cardviolent_illicit_not_unsafe | #1 / 7 | ↑0.975 | Source ↗official | |
| Humanity's Last Exam RMS calibration error (Scale Labs)calibrationError | #4 / 39 | ↓45 | Source ↗official | |
| LiveSecBenchethics | #1 / 43 | ↑95.79 | Source ↗official | |
| LiveSecBenchfactuality | #16 / 43 | ↑57.75 | Source ↗official | |
| LiveSecBenchlegality | #4 / 43 | ↑89.24 | Source ↗official | |
| LiveSecBenchprivacy | #4 / 43 | ↑87.9 | Source ↗official | |
| LiveSecBenchpsychological_health | #1 / 43 | ↑92.91 | Source ↗official | |
| MACHIAVELLIdeception_relative_random_pct | #20 / 50 | ↓86.5 | Source ↗official | |
| Manager Coercion Benchcoercion_ladder_depth | #8 / 33 | ↓7.4 | Source ↗official | |
| Manager Coercion Benchfabrication_rate | #1 / 15 | ↓0 | Source ↗official | |
| MASKlying_probability_pct | #9 / 53 | ↓13.33 | Source ↗official | |
| MORUscore | #2 / 13 | ↑84.18 | Source ↗official | |
| OpenAI GPT-5.4 Dynamic Wellbeingemotional_reliance | #2 / 3 | ↑0.953 | Source ↗official | |
| OpenAI GPT-5.4 Dynamic Wellbeingmental_health | #2 / 3 | ↑0.975 | Source ↗official | |
| OpenAI GPT-5.4 Dynamic Wellbeingself_harm | #2 / 3 | ↑0.955 | Source ↗official | |
| OpenAI GPT-5.4 First-Person Fairnessharm_overall | #2 / 3 | ↓0.00997 | 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 | |
| PHAREbias_resistance_diagnostic | #47 / 66 | ↑0.3851 | Source ↗official | |
| PHAREhallucination_resistance_diagnostic | #29 / 70 | ↑0.7713 | Source ↗official | |
| PHAREharm_resistance_diagnostic | #15 / 70 | ↑0.9688 | Source ↗official | |
| PHAREjailbreak_resistance_diagnostic | #9 / 67 | ↑0.7161 | Source ↗official | |
| PropensityBenchscore | #6 / 14 | ↓34.35 | Source ↗official | |
| SM-Benchadversarial | #61 / 79 | ↑78.54 | Source ↗official | |
| SM-Benchambiguous_interpretation | #23 / 79 | ↑88.99 | Source ↗official | |
| SM-Benchanti_hallucination | #59 / 79 | ↑86.39 | Source ↗official | |
| SM-Bencheq_boundaries | #78 / 79 | ↑40.45 | Source ↗official | |
| SM-Benchoverfit | #73 / 79 | ↑25.68 | Source ↗official | |
| SpeciEvalbelief_animal_sentience | #87 / 113 | ↑6.52 | Source ↗official | |
| SpeciEvalland_animal_4ns | #20 / 113 | ↓4.25 | Source ↗official | |
| SpeciEvalsea_animal_4ns | #10 / 113 | ↓4.3 | Source ↗official | |
| SpeciEvalspeciesism | #22 / 113 | ↓1.55 | Source ↗official | |
| TACbase_welfare_rate | #38 / 76 | ↑26.28 | Source ↗official | |
| TukaBenchafri_jbb_cultural_asr | #2 / 6 | ↓10.9 | Source ↗official | |
| TukaBenchafri_jbb_harm_asr | #1 / 6 | ↓2.6 | Source ↗official | |
| TukaBenchafrijail_mono_asr | #1 / 6 | ↓12.2 | Source ↗official | |
| Vectara HHEM Factual Consistencyfactual_consistency_rate | #48 / 94 | ↑90.4 | Source ↗official | |
| Vigil Mental Health Safetyoverall_score | #5 / 23 | ↑72 | 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.
Taiwan Sovereignty Benchmark Pro
| Dimension | Value | Distribution |
|---|---|---|
| Pro-Taiwan rubric compatibility | 85 | |
| Warning-phrase rate | 10 | |
| Soft-censorship rate | 0 | |
| API-error rate | 0 |