Model profile
Phi 3.5 Mini Instruct
Evidence summary
Phi 3.5 Mini Instruct has an estimated overall rank of #229; its 90% source-sensitivity interval is #75–#270. Its behavior-only rank is #230; company governance moves the combined estimate to #229. Published evidence spans 5 evals and 6 of 7 behavior components. Its strongest relative result is Open LLM Safety Index (jailbreakbench_safety_rate, #4 of 21); its weakest is Large-scale Moral Machine experiment on LLMs (human_choice_distance, #36 of 39).
Compare this model
Only models sharing at least one published sub-eval are listed.
Official and reference links
- Hugging Face ↗microsoft/Phi-3.5-mini-instruct
- 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 | Source |
|---|---|---|---|---|
| AILuminate General Purpose AI Chatchild_sexual_exploitation | #23 / 32 | ↓26.1 | Source ↗official | |
| AILuminate General Purpose AI Chatdefamation | #9 / 32 | ↓10.3 | Source ↗official | |
| AILuminate General Purpose AI Chathate | #10 / 32 | ↓6.2 | Source ↗official | |
| AILuminate General Purpose AI Chatindiscriminate_weapons_cbrne | #17 / 32 | ↓24.9 | Source ↗official | |
| AILuminate General Purpose AI Chatintellectual_property | #9 / 32 | ↓7.1 | Source ↗official | |
| AILuminate General Purpose AI Chatnon_violent_crimes | #12 / 32 | ↓14.1 | Source ↗official | |
| AILuminate General Purpose AI Chatprivacy | #14 / 32 | ↓11 | Source ↗official | |
| AILuminate General Purpose AI Chatsex_related_crimes | #17 / 32 | ↓15 | Source ↗official | |
| AILuminate General Purpose AI Chatsexual_content | #17 / 32 | ↓23.6 | Source ↗official | |
| AILuminate General Purpose AI Chatspecialized_advice | #15 / 32 | ↓15.1 | Source ↗official | |
| AILuminate General Purpose AI Chatsuicide_self_harm | #19 / 32 | ↓14.2 | Source ↗official | |
| AILuminate General Purpose AI Chatviolent_crimes | #16 / 32 | ↓17.6 | Source ↗official | |
| Enkrypt AI Safety Leaderboardbias_attack_non_success_rate | #193 / 241 | ↑11.37 | Source ↗official | |
| Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate | #47 / 241 | ↑92 | Source ↗official | |
| Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate | #86 / 241 | ↑79.44 | Source ↗official | |
| Enkrypt AI Safety Leaderboardtoxicity_attack_non_success_rate | #88 / 239 | ↑97.45 | Source ↗official | |
| Large-scale Moral Machine experiment on LLMshuman_choice_distance | #36 / 39 | ↓1.557 | Source ↗official | |
| Microsoft Phi Safety Panelsharmful_continuation | #3 / 10 | ↓0.006 | Source ↗official | |
| Microsoft Phi Safety Panelsharmful_summarization | #7 / 10 | ↓0.119 | Source ↗official | |
| Microsoft Phi Safety Panelsjailbreak | #5 / 10 | ↓0.119 | Source ↗official | |
| Microsoft Phi Safety Panelsthird_party_harm | #4 / 10 | ↓0.243 | Source ↗official | |
| Open LLM Safety Indexjailbreakbench_safety_rate | #4 / 21 | ↑0.6 | Source ↗official | |
| Open LLM Safety Indexstrongreject_safety_rate | #8 / 21 | ↑0.3333 | 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.