Model profile
Gemma 2 9B IT
Evidence summary
Safety. Gemma 2 9B IT has an estimated Safety rank of #227; its 90% source-sensitivity interval is #104–#259. Its behavior-only rank is #238; company governance moves the combined estimate to #227. Published Safety evidence spans 9 eval lineages and 7 of 7 components. Its strongest relative result is Enkrypt AI Safety Leaderboard (cbrn_attack_non_success_rate, #17 of 248); its weakest is Large-scale Moral Machine experiment on LLMs (human_choice_distance, #38 of 39).
Freedom. Gemma 2 9B IT has an estimated Freedom rank of #563; its 90% source-sensitivity interval is #374–#606. Published Freedom evidence spans 8 eval lineages and 1 of 1 components. Its strongest relative result is Enkrypt AI Safety Leaderboard (toxicity_attack_non_success_rate, #9 of 246); its weakest is UGI Leaderboard — base-model willingness (willingness_direct_score, #146 of 156).
Compare this model
Only models sharing at least one published sub-eval are listed.
Official and reference links
- Hugging Face ↗google/gemma-2-9b-it
- OpenRouter ↗google/gemma-2-9b-it
- Official model page ↗Exact model document · Reviewed official Hugging Face owner · official repository
- Release source ↗direct research preview identity date
Finetunes
Direct finetunes linked to this canonical base model. Quantized and repackaged derivatives are excluded.
- temaq-org/Tema_Q-X ↗Direct non-quantized finetune documented by Hugging Face metadata
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-eval | Rank | Value | Distribution | Source |
|---|---|---|---|---|
| AILuminate General Purpose AI Chatchild_sexual_exploitation | #4 / 32 | ↓6.7 | Source ↗official | |
| AILuminate General Purpose AI Chatdefamation | #7 / 32 | ↓8.5 | Source ↗official | |
| AILuminate General Purpose AI Chathate | #7 / 32 | ↓3.5 | Source ↗official | |
| AILuminate General Purpose AI Chatindiscriminate_weapons_cbrne | #8 / 32 | ↓12.8 | Source ↗official | |
| AILuminate General Purpose AI Chatintellectual_property | #8 / 32 | ↓5.8 | Source ↗official | |
| AILuminate General Purpose AI Chatnon_violent_crimes | #8 / 32 | ↓8.8 | Source ↗official | |
| AILuminate General Purpose AI Chatprivacy | #7 / 32 | ↓6.9 | Source ↗official | |
| AILuminate General Purpose AI Chatsex_related_crimes | #5 / 32 | ↓6.5 | Source ↗official | |
| AILuminate General Purpose AI Chatsexual_content | #6 / 32 | ↓6.4 | Source ↗official | |
| AILuminate General Purpose AI Chatspecialized_advice | #5 / 32 | ↓8 | Source ↗official | |
| AILuminate General Purpose AI Chatsuicide_self_harm | #6 / 32 | ↓5.8 | Source ↗official | |
| AILuminate General Purpose AI Chatviolent_crimes | #7 / 32 | ↓10.2 | Source ↗official | |
| COMPL-AI AI-Identity Disclosurescore | #5 / 14 | ↑0.9863 | Source ↗official | |
| COMPL-AI LLM RuLES Multi-Turn Rule Followingscore | #6 / 14 | ↑0.4996 | Source ↗official | |
| COMPL-AI TensorTrust Goal-Hijacking Resistancescore | #7 / 13 | ↑0.4858 | Source ↗official | |
| DSPSafeBenchscore | #8 / 12 | ↑72.34 | Source ↗official | |
| Enkrypt AI Safety Leaderboardbias_attack_non_success_rate | #123 / 248 | ↑16.02 | Source ↗official | |
| Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate | #17 / 248 | ↑95.17 | Source ↗official | |
| Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate | #90 / 248 | ↑81.11 | Source ↗official | |
| Enkrypt AI Safety Leaderboardtoxicity_attack_non_success_rate | #238 / 246 | ↑70 | Source ↗official | |
| IndoBias-Pairs — parity-aware culturally grounded biasparity_score | #12 / 26 | ↑85.94 | Source ↗official | |
| Large-scale Moral Machine experiment on LLMshuman_choice_distance | #38 / 39 | ↓1.869 | Source ↗official | |
| OR-Benchover_refusal_rate | #16 / 25 | ↓79.9 | Source ↗official | |
| OR-Benchtoxic_acceptance_rate | #9 / 25 | ↓1.5 | Source ↗official | |
| SORRY-Benchavg | #8 / 51 | ↓0.1 | Source ↗official | |
| SYCON Benchfalse_presupposition_tof | #10 / 11 | ↑1.86 | Source ↗official | |
| SYCON Benchunethical_queries_tof | #3 / 11 | ↑2.36 | Source ↗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-eval | Rank | Value | Distribution | Source |
|---|---|---|---|---|
| AILuminate General Purpose AI Chatchild_sexual_exploitation | #29 / 32 | ↑6.7 | Source ↗official | |
| AILuminate General Purpose AI Chatdefamation | #26 / 32 | ↑8.5 | Source ↗official | |
| AILuminate General Purpose AI Chathate | #26 / 32 | ↑3.5 | Source ↗official | |
| AILuminate General Purpose AI Chatindiscriminate_weapons_cbrne | #25 / 32 | ↑12.8 | Source ↗official | |
| AILuminate General Purpose AI Chatintellectual_property | #25 / 32 | ↑5.8 | Source ↗official | |
| AILuminate General Purpose AI Chatnon_violent_crimes | #25 / 32 | ↑8.8 | Source ↗official | |
| AILuminate General Purpose AI Chatprivacy | #26 / 32 | ↑6.9 | Source ↗official | |
| AILuminate General Purpose AI Chatsex_related_crimes | #28 / 32 | ↑6.5 | Source ↗official | |
| AILuminate General Purpose AI Chatsexual_content | #27 / 32 | ↑6.4 | Source ↗official | |
| AILuminate General Purpose AI Chatspecialized_advice | #27 / 32 | ↑8 | Source ↗official | |
| AILuminate General Purpose AI Chatsuicide_self_harm | #27 / 32 | ↑5.8 | Source ↗official | |
| AILuminate General Purpose AI Chatviolent_crimes | #26 / 32 | ↑10.2 | Source ↗official | |
| COMPL-AI LLM RuLES Multi-Turn Rule Followingscore | #9 / 14 | ↓0.4996 | Source ↗official | |
| DSPSafeBenchscore | #5 / 12 | ↓72.34 | Source ↗official | |
| Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate | #230 / 248 | ↓95.17 | Source ↗official | |
| Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate | #159 / 248 | ↓81.11 | Source ↗official | |
| Enkrypt AI Safety Leaderboardtoxicity_attack_non_success_rate | #9 / 246 | ↓70 | Source ↗official | |
| OR-Benchover_refusal_rate | #16 / 25 | ↓79.9 | Source ↗official | |
| OR-Benchtoxic_acceptance_rate | #17 / 25 | ↑1.5 | Source ↗official | |
| SORRY-Benchavg | #44 / 51 | ↑0.1 | Source ↗official | |
| SpeechMap model completioncomplete_pct | #169 / 181 | ↑30 | Source ↗official | |
| UGI Leaderboard — base-model willingnesswillingness_adherence_score | #126 / 156 | ↑1 | Source ↗official | |
| UGI Leaderboard — base-model willingnesswillingness_direct_score | #146 / 156 | ↑1 | 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.