Model profile
Mistral Large 3
Evidence summary
Mistral Large 3 has an estimated overall rank of #247; its 90% source-sensitivity interval is #140–#262. Its behavior-only rank is #241; company governance moves the combined estimate to #247. Published evidence spans 9 evals and 6 of 7 behavior components. Its strongest relative result is HUMAINE Trust, Ethics and Safety (trust_ethics_safety_score, #3 of 54); its weakest is Alignment Leaderboard (safety, #24 of 24).
Compare this model
Only models sharing at least one published sub-eval are listed.
Official and reference links
- Artificial Analysis ↗mistral-large-3
- Hugging Face ↗mistralai/Mistral-Large-3-675B-Instruct-2512
- 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 | #183 / 311 | 0.8375 | ↓ lower | Source ↗official | |
| Alignment Leaderboardcorrigibility | #24 / 24 | 3.845 | ↑ higher | Source ↗official | |
| Alignment Leaderboardhonesty | #23 / 24 | 3.083 | ↑ higher | Source ↗official | |
| Alignment Leaderboardnon_manipulation | #24 / 24 | 2.402 | ↑ higher | Source ↗official | |
| Alignment Leaderboardrobustness | #24 / 24 | 2.28 | ↑ higher | Source ↗official | |
| Alignment Leaderboardsafety | #24 / 24 | 2.865 | ↑ higher | Source ↗official | |
| Alignment Leaderboardscheming | #24 / 24 | 3.211 | ↑ higher | Source ↗official | |
| BullshitBench v2clear_pushback_rate | #104 / 105 | 0.02 | ↑ higher | Source ↗official | |
| Cisco AI Defense Rolling Single-Turn Leaderboardsingle_turn_attack_success_rate | #102 / 105 | 80.16 | ↓ lower | Source ↗official | |
| DystopiaBenchbasaglia_score | #41 / 50 | 69.8 | ↓ lower | Source ↗official | |
| DystopiaBenchbaudrillard_score | #47 / 50 | 77.73 | ↓ lower | Source ↗official | |
| DystopiaBenchhuxley_score | #38 / 50 | 75.8 | ↓ lower | Source ↗official | |
| DystopiaBenchlaguardia_score | #36 / 50 | 69 | ↓ lower | Source ↗official | |
| DystopiaBenchorwell_score | #29 / 50 | 70.93 | ↓ lower | Source ↗official | |
| DystopiaBenchpetrov_score | #41 / 50 | 78.47 | ↓ lower | Source ↗official | |
| HUMAINE Trust, Ethics and Safetytrust_ethics_safety_score | #3 / 54 | 29.36 | ↑ higher | Source ↗official | |
| LiveSecBenchethics | #34 / 43 | 24.59 | ↑ higher | Source ↗official | |
| LiveSecBenchfactuality | #24 / 43 | 43.2 | ↑ higher | Source ↗official | |
| LiveSecBenchlegality | #33 / 43 | 20.79 | ↑ higher | Source ↗official | |
| LiveSecBenchprivacy | #41 / 43 | 10.14 | ↑ higher | Source ↗official | |
| LiveSecBenchpsychological_health | #28 / 43 | 43.44 | ↑ higher | Source ↗official | |
| PHAREbias_resistance_diagnostic | #10 / 66 | 0.6272 | ↑ higher | Source ↗official | |
| PHAREhallucination_resistance_diagnostic | #51 / 70 | 0.6803 | ↑ higher | Source ↗official | |
| PHAREharm_resistance_diagnostic | #54 / 70 | 0.8806 | ↑ higher | Source ↗official | |
| PHAREjailbreak_resistance_diagnostic | #67 / 67 | 0.249 | ↑ higher | Source ↗official | |
| RefusalBenchyouden_j | #19 / 19 | 0.00906 | ↑ higher | 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.
UGI Political Values
Taiwan Sovereignty Benchmark Pro
| Dimension | Value | Distribution |
|---|---|---|
| Pro-Taiwan rubric compatibility | 45 | |
| Warning-phrase rate | 30 | |
| Soft-censorship rate | 5 | |
| API-error rate | 0 |