Model profile
Gemma 3 27B IT
Evidence summary
Safety. Gemma 3 27B IT has an estimated Safety rank of #265; its 90% source-sensitivity interval is #49–#294. Its behavior-only rank is #279; company governance moves the combined estimate to #265. Published Safety evidence spans 13 eval lineages and 6 of 7 components. Its strongest relative result is Enkrypt AI Safety Leaderboard (cbrn_attack_non_success_rate, #77 of 248); its weakest is Confabulations (confabulation_rate, #52 of 52).
Freedom. Gemma 3 27B IT has an estimated Freedom rank of #67; its 90% source-sensitivity interval is #69–#232. Published Freedom evidence spans 8 eval lineages and 1 of 1 components. Its strongest relative result is AILuminate General Purpose AI Chat (intellectual_property, #3 of 32); its weakest is UGI Leaderboard — base-model willingness (willingness_adherence_score, #151 of 156).
Compare this model
Only models sharing at least one published sub-eval are listed.
Official and reference links
- Artificial Analysis ↗gemma-3-27b
- Hugging Face ↗google/gemma-3-27b-it
- OpenRouter ↗google/gemma-3-27b-it
- Official model page ↗Exact model document · Reviewed official Hugging Face owner · official repository
- Release source ↗curated outward evidence
Finetunes
Direct finetunes linked to this canonical base model. Quantized and repackaged derivatives are excluded.
- Nabbers1999/Gemma-3-27B-it-NP-Abliterated ↗Direct non-quantized finetune documented by Hugging Face metadata
- soob3123/amoral-gemma3-27B-v2 ↗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 |
|---|---|---|---|---|
| AA-Omnisciencehallucination_rate | #298 / 345 | ↓0.9209 | Source ↗official | |
| AgentDrive Safety Compliancescr | #33 / 48 | ↑66.25 | Source ↗official | |
| AILuminate General Purpose AI Chatchild_sexual_exploitation | #24 / 32 | ↓27 | Source ↗official | |
| AILuminate General Purpose AI Chatdefamation | #27 / 32 | ↓38 | Source ↗official | |
| AILuminate General Purpose AI Chathate | #27 / 32 | ↓23 | Source ↗official | |
| AILuminate General Purpose AI Chatindiscriminate_weapons_cbrne | #25 / 32 | ↓42 | Source ↗official | |
| AILuminate General Purpose AI Chatintellectual_property | #30 / 32 | ↓38 | Source ↗official | |
| AILuminate General Purpose AI Chatnon_violent_crimes | #27 / 32 | ↓35 | Source ↗official | |
| AILuminate General Purpose AI Chatprivacy | #28 / 32 | ↓31 | Source ↗official | |
| AILuminate General Purpose AI Chatsex_related_crimes | #25 / 32 | ↓24 | Source ↗official | |
| AILuminate General Purpose AI Chatsexual_content | #19 / 32 | ↓27 | Source ↗official | |
| AILuminate General Purpose AI Chatspecialized_advice | #27 / 32 | ↓30 | Source ↗official | |
| AILuminate General Purpose AI Chatsuicide_self_harm | #28 / 32 | ↓29 | Source ↗official | |
| AILuminate General Purpose AI Chatviolent_crimes | #26 / 32 | ↓33 | Source ↗official | |
| Arena Factuality — Text Arena (factuality-only weighting)factuality_bt_rating | #39 / 111 | ↑1445.0 | Source ↗official | |
| BullshitBench v2clear_pushback_rate | #115 / 117 | ↑0.03 | Source ↗official | |
| Cisco AI Defense Rolling Single-Turn Leaderboardsingle_turn_attack_success_rate | #86 / 104 | ↓69.44 | Source ↗official | |
| Confabulationsconfabulation_rate | #52 / 52 | ↓66.34 | Source ↗official | |
| Enkrypt AI Safety Leaderboardbias_attack_non_success_rate | #110 / 248 | ↑17.05 | Source ↗official | |
| Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate | #77 / 248 | ↑90.33 | Source ↗official | |
| Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate | #162 / 248 | ↑56.11 | Source ↗official | |
| Enkrypt AI Safety Leaderboardtoxicity_attack_non_success_rate | #95 / 246 | ↑97.36 | Source ↗official | |
| HUMAINE Trust, Ethics and Safetytrust_ethics_safety_score | #17 / 54 | ↑27.74 | Source ↗official | |
| PHAREbias_resistance_diagnostic | #50 / 66 | ↑0.3801 | Source ↗official | |
| PHAREhallucination_resistance_diagnostic | #67 / 70 | ↑0.603 | Source ↗official | |
| PHAREharm_resistance_diagnostic | #45 / 70 | ↑0.9136 | Source ↗official | |
| PHAREjailbreak_resistance_diagnostic | #51 / 67 | ↑0.4007 | Source ↗official | |
| SOSBenchbiology_pvr | #17 / 23 | ↓0.792 | Source ↗official | |
| SOSBenchchemistry_pvr | #19 / 23 | ↓0.646 | Source ↗official | |
| SOSBenchmedicine_pvr | #19 / 23 | ↓0.814 | Source ↗official | |
| SOSBenchpharmacology_pvr | #18 / 23 | ↓0.934 | Source ↗official | |
| SOSBenchphysics_pvr | #21 / 23 | ↓0.842 | Source ↗official | |
| SOSBenchpsychology_pvr | #18 / 23 | ↓0.792 | Source ↗official | |
| Vectara HHEM Factual Consistencyfactual_consistency_rate | #31 / 94 | ↑92.6 | Source ↗official | |
| VETO Misfired Alignmentmisfired_alignment_rate_pct | #8 / 25 | ↓6.3 | 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 | #9 / 32 | ↑27 | Source ↗official | |
| AILuminate General Purpose AI Chatdefamation | #6 / 32 | ↑38 | Source ↗official | |
| AILuminate General Purpose AI Chathate | #6 / 32 | ↑23 | Source ↗official | |
| AILuminate General Purpose AI Chatindiscriminate_weapons_cbrne | #8 / 32 | ↑42 | Source ↗official | |
| AILuminate General Purpose AI Chatintellectual_property | #3 / 32 | ↑38 | Source ↗official | |
| AILuminate General Purpose AI Chatnon_violent_crimes | #6 / 32 | ↑35 | Source ↗official | |
| AILuminate General Purpose AI Chatprivacy | #5 / 32 | ↑31 | Source ↗official | |
| AILuminate General Purpose AI Chatsex_related_crimes | #7 / 32 | ↑24 | Source ↗official | |
| AILuminate General Purpose AI Chatsexual_content | #14 / 32 | ↑27 | Source ↗official | |
| AILuminate General Purpose AI Chatspecialized_advice | #6 / 32 | ↑30 | Source ↗official | |
| AILuminate General Purpose AI Chatsuicide_self_harm | #5 / 32 | ↑29 | Source ↗official | |
| AILuminate General Purpose AI Chatviolent_crimes | #7 / 32 | ↑33 | Source ↗official | |
| Cisco AI Defense Rolling Single-Turn Leaderboardsingle_turn_attack_success_rate | #19 / 104 | ↑69.44 | Source ↗official | |
| Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate | #167 / 248 | ↓90.33 | Source ↗official | |
| Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate | #87 / 248 | ↓56.11 | Source ↗official | |
| Enkrypt AI Safety Leaderboardtoxicity_attack_non_success_rate | #152 / 246 | ↓97.36 | Source ↗official | |
| PHAREharm_resistance_diagnostic | #26 / 70 | ↓0.9136 | Source ↗official | |
| PHAREjailbreak_resistance_diagnostic | #17 / 67 | ↓0.4007 | Source ↗official | |
| SOSBenchbiology_pvr | #7 / 23 | ↑0.792 | Source ↗official | |
| SOSBenchchemistry_pvr | #5 / 23 | ↑0.646 | Source ↗official | |
| SOSBenchmedicine_pvr | #5 / 23 | ↑0.814 | Source ↗official | |
| SOSBenchpharmacology_pvr | #5 / 23 | ↑0.934 | Source ↗official | |
| SOSBenchphysics_pvr | #3 / 23 | ↑0.842 | Source ↗official | |
| SOSBenchpsychology_pvr | #6 / 23 | ↑0.792 | Source ↗official | |
| SpeechMap model completioncomplete_pct | #46 / 181 | ↑74.5 | Source ↗official | |
| UGI Leaderboard — base-model willingnesswillingness_adherence_score | #151 / 156 | ↑0 | Source ↗official | |
| UGI Leaderboard — base-model willingnesswillingness_direct_score | #46 / 156 | ↑4 | Source ↗official | |
| VETO Misfired Alignmentmisfired_alignment_rate_pct | #8 / 25 | ↓6.3 | 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.