Model profile
Phi 2
Microsoftdeveloper
#195 / 267overall rank
3eval lineages
Evidence summary
Phi 2 has an estimated overall rank of #195; its 90% source-sensitivity interval is #79–#261. Its behavior-only rank is #197; company governance moves the combined estimate to #195. Published evidence spans 3 evals and 3 of 7 behavior components. Its strongest relative result is Enkrypt AI Safety Leaderboard (bias_attack_non_success_rate, #47 of 260); its weakest is Microsoft Phi Safety Panels (harmful_continuation, #10 of 10).
Compare this model
Only models sharing at least one published sub-eval are listed.
Official and reference links
- Microsoft Phi-2 ↗Exact model document · huggingface · official repository
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 |
|---|---|---|---|---|---|
| Enkrypt AI Safety Leaderboardbias_attack_non_success_rate | #47 / 260 | 31.52 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate | #178 / 260 | 85.33 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate | #258 / 260 | 21.67 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardtoxicity_attack_non_success_rate | #256 / 258 | 56.36 | ↑ higher | Source ↗official | |
| Microsoft Phi Safety Panelsharmful_continuation | #10 / 10 | 0.029 | ↓ lower | Source ↗official | |
| Microsoft Phi Safety Panelsharmful_summarization | #9 / 10 | 0.144 | ↓ lower | Source ↗official | |
| Microsoft Phi Safety Panelsjailbreak | #8 / 10 | 0.15 | ↓ lower | Source ↗official | |
| Microsoft Phi Safety Panelsthird_party_harm | #2 / 10 | 0.24 | ↓ lower | Source ↗official | |
| Open LLM Safety Indexjailbreakbench_safety_rate | #16 / 21 | 0.1333 | ↑ higher | Source ↗official | |
| Open LLM Safety Indexstrongreject_safety_rate | #19 / 21 | 0 | ↑ higher | Source ↗official |