Model profile
GPT 5.4 Mini
Evidence summary
GPT 5.4 Mini has an estimated overall rank of #168; its 90% source-sensitivity interval is #54–#218. Its behavior-only rank is #178; company governance moves the combined estimate to #168. Published evidence spans 14 evals and 6 of 7 behavior components. Its strongest relative result is DystopiaBench (petrov_score, #9 of 50); its weakest is SM-Bench (overfit, #71 of 73).
Compare this model
Only models sharing at least one published sub-eval are listed.
Official and reference links
- Artificial Analysis ↗gpt-5-4-mini
- OpenRouter ↗openai/gpt-5.4-mini
- 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 | Better | Source |
|---|---|---|---|---|---|
| AA-Omnisciencehallucination_rate | #244 / 311 | 0.8916 | ↓ lower | Source ↗official | |
| BullshitBench v2clear_pushback_rate | #49 / 105 | 0.3167 | ↑ higher | Source ↗official | |
| CAIS Risk Indexagent_red_teaming | #21 / 43 | 74.3 | ↓ lower | Source ↗official | |
| CAIS Risk Indexbioweapons_assistance | #12 / 48 | 60 | ↓ lower | Source ↗official | |
| CAIS Risk Indexhle_overconfidence | #23 / 49 | 55.2 | ↓ lower | Source ↗official | |
| CAIS Risk Indexmachiavelli | #39 / 45 | 95.6 | ↓ lower | Source ↗official | |
| CAIS Risk Indexmask | #21 / 51 | 15.9 | ↓ lower | Source ↗official | |
| CAIS Risk Indexpolitical_manipulation | #30 / 32 | 61.1 | ↓ lower | Source ↗official | |
| CAIS Risk Indextextquests_harm | #12 / 48 | 15.4 | ↓ lower | Source ↗official | |
| Cisco AI Defense Rolling Single-Turn Leaderboardsingle_turn_attack_success_rate | #26 / 105 | 9.571 | ↓ lower | Source ↗official | |
| DystopiaBenchbasaglia_score | #14 / 50 | 49.13 | ↓ lower | Source ↗official | |
| DystopiaBenchbaudrillard_score | #12 / 50 | 38.87 | ↓ lower | Source ↗official | |
| DystopiaBenchhuxley_score | #11 / 50 | 45.1 | ↓ lower | Source ↗official | |
| DystopiaBenchlaguardia_score | #11 / 50 | 52.8 | ↓ lower | Source ↗official | |
| DystopiaBenchorwell_score | #12 / 50 | 41.8 | ↓ lower | Source ↗official | |
| DystopiaBenchpetrov_score | #9 / 50 | 39.6 | ↓ lower | Source ↗official | |
| MACHIAVELLIdeception_relative_random_pct | #41 / 50 | 95.6 | ↓ lower | Source ↗official | |
| RefusalBenchyouden_j | #16 / 19 | 0.06411 | ↑ higher | Source ↗official | |
| SM-Benchadversarial | #60 / 73 | 76.83 | ↑ higher | Source ↗official | |
| SM-Benchambiguous_interpretation | #38 / 73 | 84.08 | ↑ higher | Source ↗official | |
| SM-Benchanti_hallucination | #63 / 73 | 80.23 | ↑ higher | Source ↗official | |
| SM-Bencheq_boundaries | #28 / 73 | 67.28 | ↑ higher | Source ↗official | |
| SM-Benchoverfit | #71 / 73 | 15.3 | ↑ higher | Source ↗official | |
| VETO Misfired Alignmentmisfired_alignment_rate_pct | #17 / 25 | 9.9 | ↓ 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.
Agent-ValueBench Moral Foundations (MFT08)
Agent-ValueBench HEXACO
| Dimension | Value | Distribution |
|---|---|---|
| Openness to experience | 5.5 | |
| Honesty-humility | 5.3 | |
| Extraversion | 6.1 | |
| Agreeableness | 6.5 | |
| Conscientiousness | 7.3 |