Model profile
Mistral 7B Instruct
Mistral AIdeveloper
2023-09-27release date
#205 / 267overall rank
21eval lineages
Evidence summary
Mistral 7B Instruct has an estimated overall rank of #205; its 90% source-sensitivity interval is #139–#236. Its behavior-only rank is #195; company governance moves the combined estimate to #205. Published evidence spans 21 evals and 7 of 7 behavior components. Its strongest relative result is Enkrypt AI Safety Leaderboard (bias_attack_non_success_rate, #42 of 260); its weakest is JailBench (jailbreak_success_rate, #14 of 14).
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
Published eval results
Rank is within that sub-eval. Black marks the observed result; the grey dot marks the value implied by the global rank. Values stay on each source’s native scale.
| Eval / sub-eval | Rank | Value | Distribution | Better | Source |
|---|---|---|---|---|---|
| AA-Omnisciencehallucination_rate | #138 / 311 | 0.7896 | ↓ lower | Source ↗official | |
| AbstentionBenchanswer_unknown_f1 | #15 / 20 | 0.8549 | ↑ higher | Source ↗official | |
| AbstentionBenchfalse_premise_f1 | #11 / 20 | 0.6706 | ↑ higher | Source ↗official | |
| AbstentionBenchstale_f1 | #15 / 20 | 0.6004 | ↑ higher | Source ↗official | |
| AbstentionBenchsubjective_f1 | #15 / 20 | 0.7038 | ↑ higher | Source ↗official | |
| AbstentionBenchunderspecified_context_f1 | #7 / 20 | 0.6605 | ↑ higher | Source ↗official | |
| AbstentionBenchunderspecified_intent_f1 | #12 / 20 | 0.725 | ↑ higher | Source ↗official | |
| AIRBench 2024 Safety Scenariossafety_scenarios | #77 / 80 | 0.326 | ↑ higher | Source ↗official | |
| ChineseSafescore | #20 / 22 | 42.99 | ↑ higher | Source ↗official | |
| Contextual MoralChoicehuman_agreement | #16 / 22 | 0.35 | ↑ higher | Source ↗official | |
| DSPSafeBenchscore | #6 / 12 | 73.04 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardbias_attack_non_success_rate | #42 / 260 | 36.61 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate | #167 / 260 | 86.39 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate | #247 / 260 | 31.3 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardtoxicity_attack_non_success_rate | #170 / 258 | 94.65 | ↑ higher | Source ↗official | |
| HarmBenchdr | #24 / 28 | 46.3 | ↓ lower | Source ↗official | |
| HELM Safetyanthropic_red_team | #79 / 80 | 0.806 | ↑ higher | Source ↗official | |
| HELM Safetybbq | #78 / 80 | 0.6165 | ↑ higher | Source ↗official | |
| HELM Safetyharmbench | #79 / 80 | 0.2885 | ↑ higher | Source ↗official | |
| HELM Safetysimple_safety_tests | #79 / 80 | 0.6215 | ↑ higher | Source ↗official | |
| HELM Safetyxstest | #77 / 80 | 0.8065 | ↑ higher | Source ↗official | |
| JailBenchjailbreak_success_rate | #14 / 14 | 78.14 | ↓ lower | Source ↗official | |
| Large-scale Moral Machine experiment on LLMshuman_choice_distance | #14 / 39 | 0.9189 | ↓ lower | Source ↗official | |
| Microsoft Phi Safety Panelsharmful_continuation | #9 / 10 | 0.026 | ↓ lower | Source ↗official | |
| Microsoft Phi Safety Panelsharmful_summarization | #10 / 10 | 0.223 | ↓ lower | Source ↗official | |
| Microsoft Phi Safety Panelsjailbreak | #9 / 10 | 0.156 | ↓ lower | Source ↗official | |
| Microsoft Phi Safety Panelsthird_party_harm | #9 / 10 | 0.562 | ↓ lower | Source ↗official | |
| S-Evalbase_en_overall | #22 / 22 | 34.2 | ↑ higher | Source ↗official | |
| SafeDialBenchaggression | #13 / 18 | 7.027 | ↑ higher | Source ↗official | |
| SafeDialBenchethics | #8 / 18 | 7.587 | ↑ higher | Source ↗official | |
| SafeDialBenchfairness | #13 / 18 | 7.187 | ↑ higher | Source ↗official | |
| SafeDialBenchlegality | #3 / 18 | 7.993 | ↑ higher | Source ↗official | |
| SafeDialBenchmorality | #7 / 18 | 7.28 | ↑ higher | Source ↗official | |
| SafeDialBenchprivacy | #6 / 18 | 7.607 | ↑ higher | Source ↗official | |
| SALAD-Benchattack_enhanced_human_autonomy_integrity | #28 / 33 | 6.035 | ↑ higher | Source ↗official | |
| SALAD-Benchattack_enhanced_information_safety_harms | #32 / 33 | 3.095 | ↑ higher | Source ↗official | |
| SALAD-Benchattack_enhanced_malicious_use | #29 / 33 | 4.65 | ↑ higher | Source ↗official | |
| SALAD-Benchattack_enhanced_misinformation_harms | #32 / 33 | 4.11 | ↑ higher | Source ↗official | |
| SALAD-Benchattack_enhanced_representation_toxicity | #29 / 33 | 4.09 | ↑ higher | Source ↗official | |
| SALAD-Benchattack_enhanced_socioeconomic_harms | #31 / 33 | 5.41 | ↑ higher | Source ↗official | |
| SALAD-Benchbase_human_autonomy_integrity | #30 / 33 | 67.21 | ↑ higher | Source ↗official | |
| SALAD-Benchbase_information_safety_harms | #29 / 33 | 78.74 | ↑ higher | Source ↗official | |
| SALAD-Benchbase_malicious_use | #29 / 33 | 57.35 | ↑ higher | Source ↗official | |
| SALAD-Benchbase_misinformation_harms | #29 / 33 | 75.97 | ↑ higher | Source ↗official | |
| SALAD-Benchbase_representation_toxicity | #29 / 33 | 74.01 | ↑ higher | Source ↗official | |
| SALAD-Benchbase_socioeconomic_harms | #29 / 33 | 74.14 | ↑ higher | Source ↗official | |
| SALAD-Benchmcq_human_autonomy_integrity | #18 / 33 | 32.92 | ↑ higher | Source ↗official | |
| SALAD-Benchmcq_information_safety_harms | #17 / 33 | 32.5 | ↑ higher | Source ↗official | |
| SALAD-Benchmcq_malicious_use | #17 / 33 | 34.36 | ↑ higher | Source ↗official | |
| SALAD-Benchmcq_misinformation_harms | #17 / 33 | 34.76 | ↑ higher | Source ↗official | |
| SALAD-Benchmcq_representation_toxicity | #17 / 33 | 31.3 | ↑ higher | Source ↗official | |
| SALAD-Benchmcq_socioeconomic_harms | #17 / 33 | 31.94 | ↑ higher | Source ↗official | |
| SORRY-Benchavg | #48 / 51 | 0.785 | ↓ lower | Source ↗official | |
| SuperCLUE Safetyinstruction_attack | #17 / 31 | 62.07 | ↑ higher | Source ↗official | |
| SuperCLUE Safetyresponsible_ai | #18 / 31 | 54.55 | ↑ higher | Source ↗official | |
| SuperCLUE Safetytraditional_safety | #25 / 31 | 68.09 | ↑ higher | Source ↗official | |
| VETO Misfired Alignmentmisfired_alignment_rate_pct | #6 / 25 | 5.7 | ↓ lower | Source ↗official | |
| XSTestsafe_full_compliance_rate | #1 / 3 | 0.984 | ↑ higher | Source ↗official | |
| XSTestunsafe_full_refusal_rate | #3 / 3 | 0.235 | ↑ higher | Source ↗official |