← Models

Model profile

Phi 4

Microsoftdeveloper
2024-12-11release date
#113 / 267overall rank
5eval lineages

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

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-evalRankValueDistributionBetterSource
AA-Omnisciencehallucination_rate#156 / 3110.8051↓ lowerSource ↗official
AILuminate General Purpose AI Chatchild_sexual_exploitation#7 / 3210.3↓ lowerSource ↗official
AILuminate General Purpose AI Chatdefamation#4 / 323.5↓ lowerSource ↗official
AILuminate General Purpose AI Chathate#4 / 321.9↓ lowerSource ↗official
AILuminate General Purpose AI Chatindiscriminate_weapons_cbrne#6 / 328.7↓ lowerSource ↗official
AILuminate General Purpose AI Chatintellectual_property#4 / 323↓ lowerSource ↗official
AILuminate General Purpose AI Chatnon_violent_crimes#5 / 325.1↓ lowerSource ↗official
AILuminate General Purpose AI Chatprivacy#4 / 322.5↓ lowerSource ↗official
AILuminate General Purpose AI Chatsex_related_crimes#4 / 325.2↓ lowerSource ↗official
AILuminate General Purpose AI Chatsexual_content#7 / 328.9↓ lowerSource ↗official
AILuminate General Purpose AI Chatspecialized_advice#5 / 328↓ lowerSource ↗official
AILuminate General Purpose AI Chatsuicide_self_harm#7 / 327.1↓ lowerSource ↗official
AILuminate General Purpose AI Chatviolent_crimes#4 / 325.9↓ lowerSource ↗official
Cisco AI Defense Rolling Single-Turn Leaderboardsingle_turn_attack_success_rate#51 / 10535.89↓ lowerSource ↗official
Confabulationsconfabulation_rate#47 / 5252.48↓ lowerSource ↗official
Enkrypt AI Safety Leaderboardbias_attack_non_success_rate#118 / 26016.8↑ higherSource ↗official
Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate#23 / 26095↑ higherSource ↗official
Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate#31 / 26096.67↑ higherSource ↗official
Enkrypt AI Safety Leaderboardtoxicity_attack_non_success_rate#41 / 25899.18↑ higherSource ↗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.

UGI Political Values

DimensionValueDistribution
Political Lean-19.7
Government48.7
Diplomacy68.5
Economy45.2
Society59

ValueCompass

DimensionValueDistribution
Universalism70.9
Self-direction52.4
Care / Harm33.1
Fairness / Cheating32
Ethical90.8