Model profile
Phi 4
Evidence summary
Phi 4 has an estimated overall rank of #113; its 90% source-sensitivity interval is #13–#230. Its behavior-only rank is #114; company governance moves the combined estimate to #113. Published evidence spans 5 evals and 5 of 7 behavior components. Its strongest relative result is Enkrypt AI Safety Leaderboard (cbrn_attack_non_success_rate, #23 of 260); its weakest is Confabulations (confabulation_rate, #47 of 52).
Compare this model
Only models sharing at least one published sub-eval are listed.
Official and reference links
- Artificial Analysis ↗phi-4
- Hugging Face ↗microsoft/phi-4
- OpenRouter ↗microsoft/phi-4
- Official model page ↗Exact model document · Reviewed official Hugging Face owner · official repository
- 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 | #156 / 311 | 0.8051 | ↓ lower | Source ↗official | |
| AILuminate General Purpose AI Chatchild_sexual_exploitation | #7 / 32 | 10.3 | ↓ lower | Source ↗official | |
| AILuminate General Purpose AI Chatdefamation | #4 / 32 | 3.5 | ↓ lower | Source ↗official | |
| AILuminate General Purpose AI Chathate | #4 / 32 | 1.9 | ↓ lower | Source ↗official | |
| AILuminate General Purpose AI Chatindiscriminate_weapons_cbrne | #6 / 32 | 8.7 | ↓ lower | Source ↗official | |
| AILuminate General Purpose AI Chatintellectual_property | #4 / 32 | 3 | ↓ lower | Source ↗official | |
| AILuminate General Purpose AI Chatnon_violent_crimes | #5 / 32 | 5.1 | ↓ lower | Source ↗official | |
| AILuminate General Purpose AI Chatprivacy | #4 / 32 | 2.5 | ↓ lower | Source ↗official | |
| AILuminate General Purpose AI Chatsex_related_crimes | #4 / 32 | 5.2 | ↓ lower | Source ↗official | |
| AILuminate General Purpose AI Chatsexual_content | #7 / 32 | 8.9 | ↓ lower | Source ↗official | |
| AILuminate General Purpose AI Chatspecialized_advice | #5 / 32 | 8 | ↓ lower | Source ↗official | |
| AILuminate General Purpose AI Chatsuicide_self_harm | #7 / 32 | 7.1 | ↓ lower | Source ↗official | |
| AILuminate General Purpose AI Chatviolent_crimes | #4 / 32 | 5.9 | ↓ lower | Source ↗official | |
| Cisco AI Defense Rolling Single-Turn Leaderboardsingle_turn_attack_success_rate | #51 / 105 | 35.89 | ↓ lower | Source ↗official | |
| Confabulationsconfabulation_rate | #47 / 52 | 52.48 | ↓ lower | Source ↗official | |
| Enkrypt AI Safety Leaderboardbias_attack_non_success_rate | #118 / 260 | 16.8 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate | #23 / 260 | 95 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate | #31 / 260 | 96.67 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardtoxicity_attack_non_success_rate | #41 / 258 | 99.18 | ↑ higher | 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.