Model profile
Gemma 3 4B
Googledeveloper
#243 / 267overall rank
3eval lineages
Evidence summary
Gemma 3 4B has an estimated overall rank of #243; its 90% source-sensitivity interval is #195–#253. Its behavior-only rank is #259; company governance moves the combined estimate to #243. Published evidence spans 3 evals and 4 of 7 behavior components. Its strongest relative result is Enkrypt AI Safety Leaderboard (cbrn_attack_non_success_rate, #125 of 260); its weakest is AA-Omniscience (hallucination_rate, #311 of 311).
Compare this model
Only models sharing at least one published sub-eval are listed.
Official and reference links
- Artificial Analysis ↗gemma-3-4b
- Official model documentation ↗Family-level model document · google · first party
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 |
|---|---|---|---|---|---|
| AA-Omnisciencehallucination_rate | #311 / 311 | 0.9883 | ↓ lower | Source ↗official | |
| Cisco AI Defense Rolling Single-Turn Leaderboardsingle_turn_attack_success_rate | #76 / 105 | 59.02 | ↓ lower | Source ↗official | |
| Enkrypt AI Safety Leaderboardbias_attack_non_success_rate | #128 / 260 | 16.02 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate | #125 / 260 | 88.5 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate | #196 / 260 | 48.89 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardtoxicity_attack_non_success_rate | #166 / 258 | 95.05 | ↑ higher | Source ↗official |