Model profile
Gemma 4 26B A4B IT
Evidence summary
Safety. Gemma 4 26B A4B IT has an estimated Safety rank of #72; its 90% source-sensitivity interval is #25–#194. Its behavior-only rank is #81; company governance moves the combined estimate to #72. Published Safety evidence spans 8 eval lineages and 5 of 7 components. Its strongest relative result is Vectara HHEM Factual Consistency (factual_consistency_rate, #11 of 94); its weakest is Enkrypt AI Safety Leaderboard (cbrn_attack_non_success_rate, #237 of 248).
Freedom. Gemma 4 26B A4B IT has an estimated Freedom rank of #348; its 90% source-sensitivity interval is #178–#492. Published Freedom evidence spans 5 eval lineages and 1 of 1 components. Its strongest relative result is Enkrypt AI Safety Leaderboard (cbrn_attack_non_success_rate, #11 of 248); its weakest is Enkrypt AI Safety Leaderboard (harmful_attack_non_success_rate, #215 of 248).
Compare this model
Only models sharing at least one published sub-eval are listed.
Official and reference links
- Artificial Analysis ↗gemma-4-26b-a4b
- Hugging Face ↗google/gemma-4-26B-A4B-it
- OpenRouter ↗google/gemma-4-26b-a4b-it
- Official model page ↗Exact model document · Reviewed official Hugging Face owner · official repository
- Release source ↗repository created at proxy
Finetunes
Direct finetunes linked to this canonical base model. Quantized and repackaged derivatives are excluded.
- coder3101/gemma-4-26B-A4B-it-heretic ↗Direct non-quantized finetune documented by Hugging Face metadata
huihui-ai/Huihui-gemma-4-26B-A4B-it-abliterated ↗Direct non-quantized finetune documented by Hugging Face metadata- huihui-ai/Huihui4-8B-A4B ↗Direct non-quantized finetune documented by Hugging Face metadata
- huihui-ai/Huihui4-8B-A4B-v2 ↗Direct non-quantized finetune documented by Hugging Face metadata
- TrevorJS/gemma-4-26B-A4B-it-uncensored ↗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 | #230 / 345 | ↓0.8644 | Source ↗official | |
| Arena Factuality — Text Arena (factuality-only weighting)factuality_bt_rating | #34 / 111 | ↑1448.0 | Source ↗official | |
| BullshitBench v2clear_pushback_rate | #83 / 117 | ↑0.16 | Source ↗official | |
| Enkrypt AI Safety Leaderboardbias_attack_non_success_rate | #172 / 248 | ↑12.92 | Source ↗official | |
| Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate | #237 / 248 | ↑63 | Source ↗official | |
| Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate | #33 / 248 | ↑95 | Source ↗official | |
| Enkrypt AI Safety Leaderboardtoxicity_attack_non_success_rate | #104 / 246 | ↑97.09 | Source ↗official | |
| MT-JailBench CrescendoXsafety_score | #9 / 21 | ↑26.58 | Source ↗official | |
| TACbase_welfare_rate | #39 / 87 | ↑28.21 | Source ↗self run | |
| Vectara HHEM Factual Consistencyfactual_consistency_rate | #11 / 94 | ↑94.8 | Source ↗official | |
| Vigil Mental Health Safetyoverall_score | #13 / 23 | ↑38 | 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 |
|---|---|---|---|---|
| Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate | #11 / 248 | ↓63 | Source ↗official | |
| Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate | #215 / 248 | ↓95 | Source ↗official | |
| Enkrypt AI Safety Leaderboardtoxicity_attack_non_success_rate | #142 / 246 | ↓97.09 | Source ↗official | |
| MT-JailBench CrescendoXsafety_score | #13 / 21 | ↓26.58 | Source ↗official | |
| SpeechMap model completioncomplete_pct | #42 / 181 | ↑76.3 | Source ↗official | |
| UGI Leaderboard — base-model willingnesswillingness_adherence_score | #79 / 156 | ↑1.5 | Source ↗official | |
| UGI Leaderboard — base-model willingnesswillingness_direct_score | #113 / 156 | ↑2 | Source ↗official | |
| Vigil Mental Health Safetyoverall_score | #10 / 23 | ↓38 | 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.