Model profile
Gemma 2 27B It
Evidence summary
Gemma 2 27B It has an estimated overall rank of #182; its 90% source-sensitivity interval is #108–#218. Its behavior-only rank is #190; company governance moves the combined estimate to #182. Published evidence spans 5 evals and 7 of 7 behavior components. Its strongest relative result is Enkrypt AI Safety Leaderboard (toxicity_attack_non_success_rate, #1 of 258); its weakest is Confabulations (confabulation_rate, #46 of 52).
Compare this model
Only models sharing at least one published sub-eval are listed.
Official and reference links
- Hugging Face ↗google/gemma-2-27b-it
- OpenRouter ↗google/gemma-2-27b-it
- Official model page ↗Exact model document · Reviewed official Hugging Face owner · official repository
- 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 | Better | Source |
|---|---|---|---|---|---|
| Confabulationsconfabulation_rate | #46 / 52 | 47.03 | ↓ lower | Source ↗official | |
| Enkrypt AI Safety Leaderboardbias_attack_non_success_rate | #46 / 260 | 31.78 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate | #103 / 260 | 89.67 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate | #119 / 260 | 73.89 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardtoxicity_attack_non_success_rate | #1 / 258 | 100 | ↑ higher | Source ↗official | |
| Large-scale Moral Machine experiment on LLMshuman_choice_distance | #27 / 39 | 1.19 | ↓ lower | Source ↗official | |
| OR-Benchover_refusal_rate | #14 / 25 | 62 | ↓ lower | Source ↗official | |
| OR-Benchtoxic_acceptance_rate | #11 / 25 | 3.2 | ↓ lower | Source ↗official | |
| SORRY-Benchavg | #9 / 51 | 0.11 | ↓ lower | Source ↗official |
Values evaluations
Descriptive values and political-framing results are separate from safety/ethics ranks. Each strip shows the evaluation’s observed model range; its endpoint labels state what lower and higher values mean.