Model profile
Mixtral 8x7B Instruct
Evidence summary
Safety. Mixtral 8x7B Instruct has an estimated Safety rank of #316; its 90% source-sensitivity interval is #231–#329. Its behavior-only rank is #308; company governance moves the combined estimate to #316. Published Safety evidence spans 13 eval lineages and 6 of 7 components. Its strongest relative result is SALAD-Bench (mcq_information_safety_harms, #6 of 33); its weakest is HELM Safety (harmbench, #77 of 80).
Freedom. Mixtral 8x7B Instruct has an estimated Freedom rank of #52; its 90% source-sensitivity interval is #61–#214. Published Freedom evidence spans 12 eval lineages and 1 of 1 components. Its strongest relative result is Human Pathogen Capabilities Test (HPCT) — overall refusal (refusal_rate_pct, #1 of 69); its weakest is HELM Safety (xstest, #61 of 80).
Compare this model
Only models sharing at least one published sub-eval are listed.
Official and reference links
- Artificial Analysis ↗mixtral-8x7b-instruct
- Hugging Face ↗mistralai/Mixtral-8x7B-Instruct-v0.1
- NVIDIA ↗qc69jvmznzxy/mixtral-8x7b-instruct
- OpenRouter ↗mistralai/mixtral-8x7b-instruct
- Official model page ↗Exact model document · Reviewed official Hugging Face owner · official repository
- Release source ↗repository created at proxy
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 |
|---|---|---|---|---|
| AIRBench 2024 Safety Scenariossafety_scenarios | #74 / 80 | ↑0.391 | Source ↗official | |
| CASE-Benchagreement_accuracy | #4 / 7 | ↑83.11 | Source ↗official | |
| COMPL-AI AI-Identity Disclosurescore | #9 / 14 | ↑0.8904 | Source ↗official | |
| COMPL-AI LLM RuLES Multi-Turn Rule Followingscore | #13 / 14 | ↑0.2561 | Source ↗official | |
| COMPL-AI TensorTrust Goal-Hijacking Resistancescore | #12 / 13 | ↑0.375 | Source ↗official | |
| Contextual MoralChoicehuman_agreement | #6 / 22 | ↑0.45 | Source ↗official | |
| Enkrypt AI Safety Leaderboardbias_attack_non_success_rate | #201 / 248 | ↑11.37 | Source ↗official | |
| Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate | #146 / 248 | ↑86.67 | Source ↗official | |
| Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate | #218 / 248 | ↑38.33 | Source ↗official | |
| Enkrypt AI Safety Leaderboardtoxicity_attack_non_success_rate | #158 / 246 | ↑95.18 | Source ↗official | |
| HarmBenchdr | #25 / 28 | ↓47.3 | Source ↗official | |
| HELM Safetyanthropic_red_team | #75 / 80 | ↑0.928 | Source ↗official | |
| HELM Safetybbq | #63 / 80 | ↑0.857 | Source ↗official | |
| HELM Safetyharmbench | #77 / 80 | ↑0.451 | Source ↗official | |
| HELM Safetysimple_safety_tests | #75 / 80 | ↑0.905 | Source ↗official | |
| HELM Safetyxstest | #61 / 80 | ↑0.931 | Source ↗official | |
| Human Pathogen Capabilities Test (HPCT) — overall refusalrefusal_rate_pct | #27 / 69 | ↑0 | Source ↗official | |
| MedSafetyBenchmedical_safety_score | #9 / 30 | ↑94 | Source ↗official | |
| SALAD-Benchattack_enhanced_human_autonomy_integrity | #19 / 33 | ↑11.85 | Source ↗official | |
| SALAD-Benchattack_enhanced_information_safety_harms | #25 / 33 | ↑5.21 | Source ↗official | |
| SALAD-Benchattack_enhanced_malicious_use | #25 / 33 | ↑7.67 | Source ↗official | |
| SALAD-Benchattack_enhanced_misinformation_harms | #18 / 33 | ↑9.54 | Source ↗official | |
| SALAD-Benchattack_enhanced_representation_toxicity | #19 / 33 | ↑10.35 | Source ↗official | |
| SALAD-Benchattack_enhanced_socioeconomic_harms | #21 / 33 | ↑9.52 | Source ↗official | |
| SALAD-Benchbase_human_autonomy_integrity | #28 / 33 | ↑76 | Source ↗official | |
| SALAD-Benchbase_information_safety_harms | #27 / 33 | ↑86.8 | Source ↗official | |
| SALAD-Benchbase_malicious_use | #28 / 33 | ↑67.65 | Source ↗official | |
| SALAD-Benchbase_misinformation_harms | #28 / 33 | ↑84.39 | Source ↗official | |
| SALAD-Benchbase_representation_toxicity | #25 / 33 | ↑82.05 | Source ↗official | |
| SALAD-Benchbase_socioeconomic_harms | #27 / 33 | ↑80.85 | Source ↗official | |
| SALAD-Benchmcq_human_autonomy_integrity | #9 / 33 | ↑51.39 | Source ↗official | |
| SALAD-Benchmcq_information_safety_harms | #6 / 33 | ↑51.94 | Source ↗official | |
| SALAD-Benchmcq_malicious_use | #9 / 33 | ↑53.27 | Source ↗official | |
| SALAD-Benchmcq_misinformation_harms | #9 / 33 | ↑52.86 | Source ↗official | |
| SALAD-Benchmcq_representation_toxicity | #9 / 33 | ↑52.08 | Source ↗official | |
| SALAD-Benchmcq_socioeconomic_harms | #8 / 33 | ↑48.89 | Source ↗official | |
| SORRY-Benchavg | #43 / 51 | ↓0.56 | 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 |
|---|---|---|---|---|
| AIRBench 2024 Safety Scenariossafety_scenarios | #7 / 80 | ↓0.391 | Source ↗official | |
| COMPL-AI LLM RuLES Multi-Turn Rule Followingscore | #2 / 14 | ↓0.2561 | Source ↗official | |
| Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate | #101 / 248 | ↓86.67 | Source ↗official | |
| Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate | #27 / 248 | ↓38.33 | Source ↗official | |
| Enkrypt AI Safety Leaderboardtoxicity_attack_non_success_rate | #89 / 246 | ↓95.18 | Source ↗official | |
| HarmBenchdr | #4 / 28 | ↑47.3 | Source ↗official | |
| HELM Safetyanthropic_red_team | #6 / 80 | ↓0.928 | Source ↗official | |
| HELM Safetyharmbench | #4 / 80 | ↓0.451 | Source ↗official | |
| HELM Safetysimple_safety_tests | #6 / 80 | ↓0.905 | Source ↗official | |
| HELM Safetyxstest | #61 / 80 | ↑0.931 | Source ↗official | |
| Human Pathogen Capabilities Test (HPCT) — overall refusalrefusal_rate_pct | #1 / 69 | ↓0 | Source ↗official | |
| MedSafetyBenchmedical_safety_score | #22 / 30 | ↓94 | Source ↗official | |
| SALAD-Benchattack_enhanced_human_autonomy_integrity | #15 / 33 | ↓11.85 | Source ↗official | |
| SALAD-Benchattack_enhanced_information_safety_harms | #7 / 33 | ↓5.21 | Source ↗official | |
| SALAD-Benchattack_enhanced_malicious_use | #9 / 33 | ↓7.67 | Source ↗official | |
| SALAD-Benchattack_enhanced_misinformation_harms | #16 / 33 | ↓9.54 | Source ↗official | |
| SALAD-Benchattack_enhanced_representation_toxicity | #15 / 33 | ↓10.35 | Source ↗official | |
| SALAD-Benchattack_enhanced_socioeconomic_harms | #13 / 33 | ↓9.52 | Source ↗official | |
| SALAD-Benchbase_human_autonomy_integrity | #6 / 33 | ↓76 | Source ↗official | |
| SALAD-Benchbase_information_safety_harms | #7 / 33 | ↓86.8 | Source ↗official | |
| SALAD-Benchbase_malicious_use | #6 / 33 | ↓67.65 | Source ↗official | |
| SALAD-Benchbase_misinformation_harms | #6 / 33 | ↓84.39 | Source ↗official | |
| SALAD-Benchbase_representation_toxicity | #9 / 33 | ↓82.05 | Source ↗official | |
| SALAD-Benchbase_socioeconomic_harms | #7 / 33 | ↓80.85 | Source ↗official | |
| SORRY-Benchavg | #8 / 51 | ↑0.56 | Source ↗official | |
| UGI Leaderboard — base-model willingnesswillingness_adherence_score | #33 / 156 | ↑5.5 | Source ↗official | |
| UGI Leaderboard — base-model willingnesswillingness_direct_score | #113 / 156 | ↑2 | 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.