← Models

Model profile

Mistral Large 3

Mistral AIdeveloper
2025-12-01release date
#309 / 333Safety rank
#2 / 645Freedom rank

Evidence summary

Safety. Mistral Large 3 has an estimated Safety rank of #309; its 90% source-sensitivity interval is #194–#325. Its behavior-only rank is #300; company governance moves the combined estimate to #309. Published Safety evidence spans 16 eval lineages and 6 of 7 components. Its strongest relative result is HUMAINE Trust, Ethics and Safety (trust_ethics_safety_score, #3 of 54); its weakest is Adversarial Humanities Benchmark (AHB) — Table 5 (non_violent_crimes, #31 of 31).

Freedom. Mistral Large 3 has an estimated Freedom rank of #2; its 90% source-sensitivity interval is #4–#63. Published Freedom evidence spans 6 eval lineages and 1 of 1 components. Its strongest relative result is PHARE (jailbreak_resistance_diagnostic, #1 of 67); its weakest is Adversarial Humanities Benchmark (AHB) — Table 5 (privacy, #12 of 31).

Compare this model

Only models sharing at least one published sub-eval are listed.

Official and reference links

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-evalRankValueDistributionSource
AA-Omnisciencehallucination_rate#226 / 3450.8596Source ↗official
Adversarial Humanities Benchmark (AHB) — Table 5child_sexual_exploitation#30 / 3180Source ↗official
Adversarial Humanities Benchmark (AHB) — Table 5defamation#23 / 3179.1Source ↗official
Adversarial Humanities Benchmark (AHB) — Table 5hate#29 / 3173.3Source ↗official
Adversarial Humanities Benchmark (AHB) — Table 5indiscriminate_weapons_cbrne#30 / 3190.5Source ↗official
Adversarial Humanities Benchmark (AHB) — Table 5intellectual_property#27 / 3172.7Source ↗official
Adversarial Humanities Benchmark (AHB) — Table 5non_violent_crimes#31 / 3197.7Source ↗official
Adversarial Humanities Benchmark (AHB) — Table 5privacy#20 / 3176.1Source ↗official
Adversarial Humanities Benchmark (AHB) — Table 5sex_related_crimes#28 / 3186.4Source ↗official
Adversarial Humanities Benchmark (AHB) — Table 5sexual_content#29 / 3167.4Source ↗official
Adversarial Humanities Benchmark (AHB) — Table 5specialized_advice#31 / 3177.1Source ↗official
Adversarial Humanities Benchmark (AHB) — Table 5suicide_self_harm#31 / 3184.4Source ↗official
Adversarial Humanities Benchmark (AHB) — Table 5violent_crimes#25 / 3183.3Source ↗official
Alignment Leaderboardcorrigibility#24 / 243.845Source ↗official
Alignment Leaderboardhonesty#23 / 243.083Source ↗official
Alignment Leaderboardnon_manipulation#24 / 242.402Source ↗official
Alignment Leaderboardrobustness#24 / 242.28Source ↗official
Alignment Leaderboardsafety#24 / 242.865Source ↗official
Alignment Leaderboardscheming#24 / 243.211Source ↗official
Arena Factuality — Text Arena (factuality-only weighting)factuality_bt_rating#79 / 1111422.0Source ↗official
AuAu Authoritarian Response Auditbehavioral_vignette_arr_percent#14 / 1728Source ↗official
AuAu Authoritarian Response Auditrealistic_prompt_arr_percent#12 / 172.7Source ↗official
BioTIERpermit_compliance_pct#8 / 5299.6Source ↗official
BioTIERrefuse_compliance_pct#49 / 529.5Source ↗official
BullshitBench v2clear_pushback_rate#116 / 1170.02Source ↗official
Cisco AI Defense Rolling Single-Turn Leaderboardsingle_turn_attack_success_rate#101 / 10480.16Source ↗official
DystopiaBenchbasaglia_score#41 / 5069.8Source ↗official
DystopiaBenchbaudrillard_score#47 / 5077.73Source ↗official
DystopiaBenchhuxley_score#38 / 5075.8Source ↗official
DystopiaBenchlaguardia_score#36 / 5069Source ↗official
DystopiaBenchorwell_score#29 / 5070.93Source ↗official
DystopiaBenchpetrov_score#41 / 5078.47Source ↗official
HUMAINE Trust, Ethics and Safetytrust_ethics_safety_score#3 / 5429.36Source ↗official
LiveSecBenchethics#34 / 4324.59Source ↗official
LiveSecBenchfactuality#24 / 4343.2Source ↗official
LiveSecBenchlegality#33 / 4320.79Source ↗official
LiveSecBenchprivacy#41 / 4310.14Source ↗official
LiveSecBenchpsychological_health#28 / 4343.44Source ↗official
PHAREbias_resistance_diagnostic#10 / 660.6272Source ↗official
PHAREhallucination_resistance_diagnostic#51 / 700.6803Source ↗official
PHAREharm_resistance_diagnostic#54 / 700.8806Source ↗official
PHAREjailbreak_resistance_diagnostic#67 / 670.249Source ↗official
RealityTest — Text AI-Identity Disclosuredisclosure_probability#17 / 170.078Source ↗official
RefusalBenchyouden_j#18 / 190.004237Source ↗official
StereoTales Harmful Associationsbenign_significant_association_score#10 / 2386.47Source ↗official
Vectara HHEM Factual Consistencyfactual_consistency_rate#83 / 9485.5Source ↗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-evalRankValueDistributionSource
Adversarial Humanities Benchmark (AHB) — Table 5child_sexual_exploitation#2 / 3180Source ↗official
Adversarial Humanities Benchmark (AHB) — Table 5defamation#9 / 3179.1Source ↗official
Adversarial Humanities Benchmark (AHB) — Table 5hate#3 / 3173.3Source ↗official
Adversarial Humanities Benchmark (AHB) — Table 5indiscriminate_weapons_cbrne#2 / 3190.5Source ↗official
Adversarial Humanities Benchmark (AHB) — Table 5intellectual_property#5 / 3172.7Source ↗official
Adversarial Humanities Benchmark (AHB) — Table 5non_violent_crimes#1 / 3197.7Source ↗official
Adversarial Humanities Benchmark (AHB) — Table 5privacy#12 / 3176.1Source ↗official
Adversarial Humanities Benchmark (AHB) — Table 5sex_related_crimes#4 / 3186.4Source ↗official
Adversarial Humanities Benchmark (AHB) — Table 5sexual_content#3 / 3167.4Source ↗official
Adversarial Humanities Benchmark (AHB) — Table 5specialized_advice#1 / 3177.1Source ↗official
Adversarial Humanities Benchmark (AHB) — Table 5suicide_self_harm#1 / 3184.4Source ↗official
Adversarial Humanities Benchmark (AHB) — Table 5violent_crimes#7 / 3183.3Source ↗official
BioTIERpermit_compliance_pct#8 / 5299.6Source ↗official
BioTIERrefuse_compliance_pct#4 / 529.5Source ↗official
Cisco AI Defense Rolling Single-Turn Leaderboardsingle_turn_attack_success_rate#4 / 10480.16Source ↗official
LiveSecBenchethics#10 / 4324.59Source ↗official
LiveSecBenchlegality#10 / 4320.79Source ↗official
LiveSecBenchprivacy#3 / 4310.14Source ↗official
LiveSecBenchpsychological_health#16 / 4343.44Source ↗official
PHAREharm_resistance_diagnostic#17 / 700.8806Source ↗official
PHAREjailbreak_resistance_diagnostic#1 / 670.249Source ↗official
UGI Leaderboard — base-model willingnesswillingness_adherence_score#4 / 1568.25Source ↗official
UGI Leaderboard — base-model willingnesswillingness_direct_score#21 / 1565Source ↗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.

UGI Political Values

DimensionValueDistribution
Political Lean-11.2
Government45.1
Diplomacy61.8
Economy43.7
Society59.8

Taiwan Sovereignty Benchmark Pro

DimensionValueDistribution
Pro-Taiwan rubric compatibility45
Warning-phrase rate30
Soft-censorship rate5
API-error rate0

The Economist World Values Survey Cultural Map

DimensionValueDistribution
Survival ↔ Self-expression0.866
Traditional ↔ Secular0.9