← Models

Model profile

Gemma 4 12B

Googledeveloper
2026-06-03release date
Not rankedSafety rank
#402 / 662Freedom rank

Evidence summary

Safety. Gemma 4 12B does not meet the evidence gate for a Safety rank. Published Safety evidence spans 2 eval lineages and 4 of 7 components. Its strongest relative result is Enkrypt AI Safety Leaderboard (harmful_attack_non_success_rate, #47 of 270); its weakest is Enkrypt AI Safety Leaderboard (cbrn_attack_non_success_rate, #244 of 270).

Freedom. Gemma 4 12B has an estimated Freedom rank of #402; its 90% source-sensitivity interval is #259–#515. Published Freedom evidence spans 1 eval lineages and 1 of 1 components. Its strongest relative result is Enkrypt AI Safety Leaderboard (cbrn_attack_non_success_rate, #27 of 270); its weakest is Enkrypt AI Safety Leaderboard (harmful_attack_non_success_rate, #224 of 270).

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#148 / 358↓0.7341Source ↗official
Enkrypt AI Safety Leaderboardbias_attack_non_success_rate#233 / 270↑10.34Source ↗official
Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate#244 / 270↑65.5Source ↗official
Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate#47 / 270↑95.56Source ↗official
Enkrypt AI Safety Leaderboardtoxicity_attack_non_success_rate#117 / 268↑96.91Source ↗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
Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate#27 / 270↓65.5Source ↗official
Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate#224 / 270↓95.56Source ↗official
Enkrypt AI Safety Leaderboardtoxicity_attack_non_success_rate#148 / 268↓96.91Source ↗official