← Models

Model profile

Gemma 3 12B

Googledeveloper
2025-03-10release date
#253 / 346Safety rank
#157 / 662Freedom rank

Evidence summary

Safety. Gemma 3 12B has an estimated Safety rank of #253; its 90% source-sensitivity interval is #89–#313. Its behavior-only rank is #263; company governance moves the combined estimate to #253. Published Safety evidence spans 9 eval lineages and 6 of 7 components. Its strongest relative result is Vectara HHEM Factual Consistency (factual_consistency_rate, #5 of 94); its weakest is PHARE (hallucination_resistance_diagnostic, #70 of 70).

Freedom. Gemma 3 12B has an estimated Freedom rank of #157; its 90% source-sensitivity interval is #80–#362. Published Freedom evidence spans 4 eval lineages and 1 of 1 components. Its strongest relative result is Cisco AI Defense Rolling Single-Turn Leaderboard (single_turn_attack_success_rate, #21 of 104); its weakest is Enkrypt AI Safety Leaderboard (toxicity_attack_non_success_rate, #196 of 268).

Compare this model

Only models sharing at least one published sub-eval are listed.

Official and reference links

Safety evals

Rank and direction are specific to the Safety portfolio. Black marks the observed result; the grey dot marks the value implied by the Safety rank. Values stay on each source’s native scale.

Eval / sub-evalRankValueDistributionSource
AA-Omnisciencehallucination_rate#355 / 358↓0.974Source ↗official
AgentDrive Safety Compliancescr#28 / 48↑77.5Source ↗official
AIMS Safety-Classifier Competenceaverage_harmful_f1#7 / 11↑0.802Source ↗official
Cisco AI Defense Rolling Single-Turn Leaderboardsingle_turn_attack_success_rate#84 / 104↓65.55Source ↗official
Enkrypt AI Safety Leaderboardbias_attack_non_success_rate#180 / 270↑13.18Source ↗official
Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate#72 / 270↑90.67Source ↗official
Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate#202 / 270↑52.78Source ↗official
Enkrypt AI Safety Leaderboardtoxicity_attack_non_success_rate#72 / 268↑98.05Source ↗official
KIDBench Implicit Child Cueimplicit_child_cue_total_mean#7 / 13↑3.98Source ↗official
PHAREbias_resistance_diagnostic#60 / 66↑0.3214Source ↗official
PHAREhallucination_resistance_diagnostic#70 / 70↑0.5386Source ↗official
PHAREharm_resistance_diagnostic#39 / 70↑0.9265Source ↗official
PHAREjailbreak_resistance_diagnostic#43 / 67↑0.4427Source ↗official
ThaiSafetyBenchsafety_score#13 / 18↑79.6Source ↗official
Vectara HHEM Factual Consistencyfactual_consistency_rate#5 / 94↑95.6Source ↗official

Freedom evals

Rank and direction are specific to the Freedom portfolio. Black marks the observed result; the grey dot marks the value implied by the Freedom rank. Values stay on each source’s native scale.

Eval / sub-evalRankValueDistributionSource
Cisco AI Defense Rolling Single-Turn Leaderboardsingle_turn_attack_success_rate#21 / 104↑65.55Source ↗official
Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate#197 / 270↓90.67Source ↗official
Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate#69 / 270↓52.78Source ↗official
Enkrypt AI Safety Leaderboardtoxicity_attack_non_success_rate#196 / 268↓98.05Source ↗official
PHAREharm_resistance_diagnostic#32 / 70↓0.9265Source ↗official
PHAREjailbreak_resistance_diagnostic#25 / 67↓0.4427Source ↗official
ThaiSafetyBenchsafety_score#6 / 18↓79.6Source ↗official