Model profile
Gemma 7B It
Googledeveloper
2024-02-21release date
#185 / 305overall rank
7eval lineages
Evidence summary
Gemma 7B It has an estimated overall rank of #185; its 90% source-sensitivity interval is #58–#237. Its behavior-only rank is #192; company governance moves the combined estimate to #185. Published evidence spans 7 evals and 7 of 7 behavior components. Its strongest relative result is SALAD-Bench (base_representation_toxicity, #6 of 33); its weakest is Enkrypt AI Safety Leaderboard (toxicity_attack_non_success_rate, #224 of 239).
Compare this model
Only models sharing at least one published sub-eval are listed.
Official and reference links
- OpenRouter ↗google/gemma-7b-it
- Official model documentation ↗Family-level model document · google · first party
- Release source ↗direct research preview identity date
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 | Source |
|---|---|---|---|---|
| Enkrypt AI Safety Leaderboardbias_attack_non_success_rate | #82 / 241 | ↑20.41 | Source ↗official | |
| Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate | #206 / 241 | ↑78.83 | Source ↗official | |
| Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate | #164 / 241 | ↑54.44 | Source ↗official | |
| Enkrypt AI Safety Leaderboardtoxicity_attack_non_success_rate | #224 / 239 | ↑77.05 | Source ↗official | |
| Large-scale Moral Machine experiment on LLMshuman_choice_distance | #32 / 39 | ↓1.314 | Source ↗official | |
| Microsoft Phi Safety Panelsharmful_continuation | #6 / 10 | ↓0.013 | Source ↗official | |
| Microsoft Phi Safety Panelsharmful_summarization | #3 / 10 | ↓0.103 | Source ↗official | |
| Microsoft Phi Safety Panelsjailbreak | #4 / 10 | ↓0.114 | Source ↗official | |
| Microsoft Phi Safety Panelsthird_party_harm | #8 / 10 | ↓0.383 | Source ↗official | |
| OR-Benchover_refusal_rate | #7 / 25 | ↓26.3 | Source ↗official | |
| OR-Benchtoxic_acceptance_rate | #17 / 25 | ↓14.5 | Source ↗official | |
| S-Evalbase_en_overall | #10 / 22 | ↑61.8 | Source ↗official | |
| SALAD-Benchattack_enhanced_human_autonomy_integrity | #16 / 33 | ↑13.36 | Source ↗official | |
| SALAD-Benchattack_enhanced_information_safety_harms | #11 / 33 | ↑22.8 | Source ↗official | |
| SALAD-Benchattack_enhanced_malicious_use | #19 / 33 | ↑9.95 | Source ↗official | |
| SALAD-Benchattack_enhanced_misinformation_harms | #12 / 33 | ↑18.91 | Source ↗official | |
| SALAD-Benchattack_enhanced_representation_toxicity | #15 / 33 | ↑17.56 | Source ↗official | |
| SALAD-Benchattack_enhanced_socioeconomic_harms | #18 / 33 | ↑12.12 | Source ↗official | |
| SALAD-Benchbase_human_autonomy_integrity | #17 / 33 | ↑94.82 | Source ↗official | |
| SALAD-Benchbase_information_safety_harms | #9 / 33 | ↑97.49 | Source ↗official | |
| SALAD-Benchbase_malicious_use | #17 / 33 | ↑93.54 | Source ↗official | |
| SALAD-Benchbase_misinformation_harms | #12 / 33 | ↑95.57 | Source ↗official | |
| SALAD-Benchbase_representation_toxicity | #6 / 33 | ↑94.42 | Source ↗official | |
| SALAD-Benchbase_socioeconomic_harms | #22 / 33 | ↑86.13 | Source ↗official | |
| SALAD-Benchmcq_human_autonomy_integrity | #15 / 33 | ↑40.56 | Source ↗official | |
| SALAD-Benchmcq_information_safety_harms | #12 / 33 | ↑40.56 | Source ↗official | |
| SALAD-Benchmcq_malicious_use | #14 / 33 | ↑40.38 | Source ↗official | |
| SALAD-Benchmcq_misinformation_harms | #16 / 33 | ↑38.1 | Source ↗official | |
| SALAD-Benchmcq_representation_toxicity | #16 / 33 | ↑38.85 | Source ↗official | |
| SALAD-Benchmcq_socioeconomic_harms | #14 / 33 | ↑38.89 | Source ↗official | |
| SORRY-Benchavg | #16 / 51 | ↓0.18 | Source ↗official |