Model profile
Mistral 7B Instruct
Mistral AIdeveloper
2023-09-27release date
#274 / 333Safety rank
#40 / 645Freedom rank
Evidence summary
Safety. Mistral 7B Instruct has an estimated Safety rank of #274; its 90% source-sensitivity interval is #199–#294. Its behavior-only rank is #262; company governance moves the combined estimate to #274. Published Safety evidence spans 24 eval lineages and 7 of 7 components. Its strongest relative result is Enkrypt AI Safety Leaderboard (bias_attack_non_success_rate, #40 of 248); its weakest is COMPL-AI TensorTrust Goal-Hijacking Resistance (score, #13 of 13).
Freedom. Mistral 7B Instruct has an estimated Freedom rank of #40; its 90% source-sensitivity interval is #48–#213. Published Freedom evidence spans 18 eval lineages and 1 of 1 components. Its strongest relative result is HELM Safety (harmbench, #2 of 80); its weakest is HELM Safety (xstest, #77 of 80).
Compare this model
Only models sharing at least one published sub-eval are listed.
Official and reference links
- Artificial Analysis ↗mistral-7b-instruct
- Hugging Face ↗mistralai/Mistral-7B-Instruct-v0.1
- OpenRouter ↗mistralai/mistral-7b-instruct
- 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.
- evolveon/Mistral-7B-Instruct-v0.3-abliterated ↗Direct non-quantized finetune documented by Hugging Face metadata
- yethdev/mistral-7b-manumit-v1 ↗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 | #177 / 345 | ↓0.8126 | Source ↗official | |
| AbstentionBenchanswer_unknown_f1 | #15 / 20 | ↑0.8549 | Source ↗official | |
| AbstentionBenchfalse_premise_f1 | #11 / 20 | ↑0.6706 | Source ↗official | |
| AbstentionBenchstale_f1 | #15 / 20 | ↑0.6004 | Source ↗official | |
| AbstentionBenchsubjective_f1 | #15 / 20 | ↑0.7038 | Source ↗official | |
| AbstentionBenchunderspecified_context_f1 | #7 / 20 | ↑0.6605 | Source ↗official | |
| AbstentionBenchunderspecified_intent_f1 | #12 / 20 | ↑0.725 | Source ↗official | |
| AIRBench 2024 Safety Scenariossafety_scenarios | #77 / 80 | ↑0.326 | Source ↗official | |
| ChineseSafescore | #20 / 22 | ↑42.99 | Source ↗official | |
| COMPL-AI AI-Identity Disclosurescore | #4 / 14 | ↑0.9932 | Source ↗official | |
| COMPL-AI LLM RuLES Multi-Turn Rule Followingscore | #14 / 14 | ↑0.1723 | Source ↗official | |
| COMPL-AI TensorTrust Goal-Hijacking Resistancescore | #13 / 13 | ↑0.3022 | Source ↗official | |
| Contextual MoralChoicehuman_agreement | #16 / 22 | ↑0.35 | Source ↗official | |
| DSPSafeBenchscore | #6 / 12 | ↑73.04 | Source ↗official | |
| Enkrypt AI Safety Leaderboardbias_attack_non_success_rate | #40 / 248 | ↑36.61 | Source ↗official | |
| Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate | #151 / 248 | ↑86.39 | Source ↗official | |
| Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate | #235 / 248 | ↑31.3 | Source ↗official | |
| Enkrypt AI Safety Leaderboardtoxicity_attack_non_success_rate | #163 / 246 | ↑94.65 | Source ↗official | |
| HarmBenchdr | #24 / 28 | ↓46.3 | Source ↗official | |
| HELM Safetyanthropic_red_team | #79 / 80 | ↑0.806 | Source ↗official | |
| HELM Safetybbq | #78 / 80 | ↑0.6165 | Source ↗official | |
| HELM Safetyharmbench | #79 / 80 | ↑0.2885 | Source ↗official | |
| HELM Safetysimple_safety_tests | #79 / 80 | ↑0.6215 | Source ↗official | |
| HELM Safetyxstest | #77 / 80 | ↑0.8065 | Source ↗official | |
| JailBenchjailbreak_success_rate | #14 / 14 | ↓78.14 | Source ↗official | |
| Large-scale Moral Machine experiment on LLMshuman_choice_distance | #14 / 39 | ↓0.9189 | Source ↗official | |
| MedSafetyBenchmedical_safety_score | #13 / 30 | ↑83.81 | Source ↗official | |
| Microsoft Phi Safety Panelsharmful_continuation | #9 / 10 | ↓0.026 | Source ↗official | |
| Microsoft Phi Safety Panelsharmful_summarization | #10 / 10 | ↓0.223 | Source ↗official | |
| Microsoft Phi Safety Panelsjailbreak | #9 / 10 | ↓0.156 | Source ↗official | |
| Microsoft Phi Safety Panelsthird_party_harm | #9 / 10 | ↓0.562 | Source ↗official | |
| NESSiE Necessary Safety Benchmarksh_score | #15 / 16 | ↑28.21 | Source ↗official | |
| S-Evalbase_en_overall | #22 / 22 | ↑34.2 | Source ↗official | |
| SafeDialBenchaggression | #13 / 18 | ↑7.027 | Source ↗official | |
| SafeDialBenchethics | #8 / 18 | ↑7.587 | Source ↗official | |
| SafeDialBenchfairness | #13 / 18 | ↑7.187 | Source ↗official | |
| SafeDialBenchlegality | #3 / 18 | ↑7.993 | Source ↗official | |
| SafeDialBenchmorality | #7 / 18 | ↑7.28 | Source ↗official | |
| SafeDialBenchprivacy | #6 / 18 | ↑7.607 | Source ↗official | |
| SALAD-Benchattack_enhanced_human_autonomy_integrity | #28 / 33 | ↑6.035 | Source ↗official | |
| SALAD-Benchattack_enhanced_information_safety_harms | #32 / 33 | ↑3.095 | Source ↗official | |
| SALAD-Benchattack_enhanced_malicious_use | #29 / 33 | ↑4.65 | Source ↗official | |
| SALAD-Benchattack_enhanced_misinformation_harms | #32 / 33 | ↑4.11 | Source ↗official | |
| SALAD-Benchattack_enhanced_representation_toxicity | #29 / 33 | ↑4.09 | Source ↗official | |
| SALAD-Benchattack_enhanced_socioeconomic_harms | #31 / 33 | ↑5.41 | Source ↗official | |
| SALAD-Benchbase_human_autonomy_integrity | #30 / 33 | ↑67.21 | Source ↗official | |
| SALAD-Benchbase_information_safety_harms | #29 / 33 | ↑78.74 | Source ↗official | |
| SALAD-Benchbase_malicious_use | #29 / 33 | ↑57.35 | Source ↗official | |
| SALAD-Benchbase_misinformation_harms | #29 / 33 | ↑75.97 | Source ↗official | |
| SALAD-Benchbase_representation_toxicity | #29 / 33 | ↑74.01 | Source ↗official | |
| SALAD-Benchbase_socioeconomic_harms | #29 / 33 | ↑74.14 | Source ↗official | |
| SALAD-Benchmcq_human_autonomy_integrity | #18 / 33 | ↑32.92 | Source ↗official | |
| SALAD-Benchmcq_information_safety_harms | #17 / 33 | ↑32.5 | Source ↗official | |
| SALAD-Benchmcq_malicious_use | #17 / 33 | ↑34.36 | Source ↗official | |
| SALAD-Benchmcq_misinformation_harms | #17 / 33 | ↑34.76 | Source ↗official | |
| SALAD-Benchmcq_representation_toxicity | #17 / 33 | ↑31.3 | Source ↗official | |
| SALAD-Benchmcq_socioeconomic_harms | #17 / 33 | ↑31.94 | Source ↗official | |
| SORRY-Benchavg | #48 / 51 | ↓0.785 | Source ↗official | |
| SuperCLUE Safetyinstruction_attack | #17 / 31 | ↑62.07 | Source ↗official | |
| SuperCLUE Safetyresponsible_ai | #18 / 31 | ↑54.55 | Source ↗official | |
| SuperCLUE Safetytraditional_safety | #25 / 31 | ↑68.09 | Source ↗official | |
| VETO Misfired Alignmentmisfired_alignment_rate_pct | #6 / 25 | ↓5.7 | Source ↗official | |
| XSTestsafe_full_compliance_rate | #1 / 3 | ↑0.984 | Source ↗official | |
| XSTestunsafe_full_refusal_rate | #3 / 3 | ↑0.235 | 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 | #4 / 80 | ↓0.326 | Source ↗official | |
| COMPL-AI LLM RuLES Multi-Turn Rule Followingscore | #1 / 14 | ↓0.1723 | Source ↗official | |
| DSPSafeBenchscore | #7 / 12 | ↓73.04 | Source ↗official | |
| Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate | #98 / 248 | ↓86.39 | Source ↗official | |
| Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate | #14 / 248 | ↓31.3 | Source ↗official | |
| Enkrypt AI Safety Leaderboardtoxicity_attack_non_success_rate | #84 / 246 | ↓94.65 | Source ↗official | |
| HarmBenchdr | #5 / 28 | ↑46.3 | Source ↗official | |
| HELM Safetyanthropic_red_team | #2 / 80 | ↓0.806 | Source ↗official | |
| HELM Safetyharmbench | #2 / 80 | ↓0.2885 | Source ↗official | |
| HELM Safetysimple_safety_tests | #2 / 80 | ↓0.6215 | Source ↗official | |
| HELM Safetyxstest | #77 / 80 | ↑0.8065 | Source ↗official | |
| JailBenchjailbreak_success_rate | #1 / 14 | ↑78.14 | Source ↗official | |
| MedSafetyBenchmedical_safety_score | #18 / 30 | ↓83.81 | Source ↗official | |
| Microsoft Phi Safety Panelsharmful_continuation | #2 / 10 | ↑0.026 | Source ↗official | |
| Microsoft Phi Safety Panelsharmful_summarization | #1 / 10 | ↑0.223 | Source ↗official | |
| Microsoft Phi Safety Panelsjailbreak | #2 / 10 | ↑0.156 | Source ↗official | |
| Microsoft Phi Safety Panelsthird_party_harm | #2 / 10 | ↑0.562 | Source ↗official | |
| S-Evalbase_en_overall | #1 / 22 | ↓34.2 | Source ↗official | |
| SafeDialBenchaggression | #6 / 18 | ↓7.027 | Source ↗official | |
| SafeDialBenchethics | #11 / 18 | ↓7.587 | Source ↗official | |
| SafeDialBenchfairness | #5 / 18 | ↓7.187 | Source ↗official | |
| SafeDialBenchlegality | #16 / 18 | ↓7.993 | Source ↗official | |
| SafeDialBenchmorality | #12 / 18 | ↓7.28 | Source ↗official | |
| SafeDialBenchprivacy | #13 / 18 | ↓7.607 | Source ↗official | |
| SALAD-Benchattack_enhanced_human_autonomy_integrity | #6 / 33 | ↓6.035 | Source ↗official | |
| SALAD-Benchattack_enhanced_information_safety_harms | #2 / 33 | ↓3.095 | Source ↗official | |
| SALAD-Benchattack_enhanced_malicious_use | #5 / 33 | ↓4.65 | Source ↗official | |
| SALAD-Benchattack_enhanced_misinformation_harms | #2 / 33 | ↓4.11 | Source ↗official | |
| SALAD-Benchattack_enhanced_representation_toxicity | #5 / 33 | ↓4.09 | Source ↗official | |
| SALAD-Benchattack_enhanced_socioeconomic_harms | #3 / 33 | ↓5.41 | Source ↗official | |
| SALAD-Benchbase_human_autonomy_integrity | #4 / 33 | ↓67.21 | Source ↗official | |
| SALAD-Benchbase_information_safety_harms | #5 / 33 | ↓78.74 | Source ↗official | |
| SALAD-Benchbase_malicious_use | #5 / 33 | ↓57.35 | Source ↗official | |
| SALAD-Benchbase_misinformation_harms | #5 / 33 | ↓75.97 | Source ↗official | |
| SALAD-Benchbase_representation_toxicity | #5 / 33 | ↓74.01 | Source ↗official | |
| SALAD-Benchbase_socioeconomic_harms | #5 / 33 | ↓74.14 | Source ↗official | |
| SORRY-Benchavg | #4 / 51 | ↑0.785 | Source ↗official | |
| SpeechMap model completioncomplete_pct | #36 / 181 | ↑78.3 | Source ↗official | |
| SuperCLUE Safetyinstruction_attack | #13 / 31 | ↓62.07 | Source ↗official | |
| SuperCLUE Safetytraditional_safety | #5 / 31 | ↓68.09 | Source ↗official | |
| VETO Misfired Alignmentmisfired_alignment_rate_pct | #6 / 25 | ↓5.7 | Source ↗official | |
| XSTestsafe_full_compliance_rate | #1 / 3 | ↑0.984 | Source ↗official | |
| XSTestunsafe_full_refusal_rate | #1 / 3 | ↓0.235 | Source ↗official |