Gemma 4 E2B

Evidence summary

Safety. Gemma 4 E2B 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 AA-Omniscience (hallucination_rate, #42 of 358); its weakest is Enkrypt AI Safety Leaderboard (bias_attack_non_success_rate, #268 of 270).

Freedom. Gemma 4 E2B has an estimated Freedom rank of #247; its 90% source-sensitivity interval is #151–#386. 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, #34 of 270); its weakest is Enkrypt AI Safety Leaderboard (harmful_attack_non_success_rate, #182 of 270).