Introducing Mistral 3 2 Dec 2025
Mistral Large 3
Mistral Large 3.
47
Benchmarks with a score
0
Ranked first, out of five or more models
0
In the top three
Mistral AI Dec 2025 Mistral Large 3
Open weights Flagship
No Epoch Capabilities Index score yet
- CritPt 0% #146 /160
- Vals Finance Agent 18.1% #45 /49
- CorpFin v2 61% #61 /119
Where it ranks
Its best score on each benchmark, and where that ranks among every model with a score there. We rank each model once, at its best setting.
Knowledge and language
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| LMArena (Elo) | 1,414 | #113 of 333 | Official leaderboard | |
| HELM Arabic | 79.0 | #12 of 28 | Official leaderboard | |
| HELM Arabic Enterprise | 66.0 | #10 of 23 | Official leaderboard | |
| MMMLU | 85.5% | #30 of 62 | Model card | |
| SimpleQA | 23.8% | #33 of 59 | Model card | |
| Judgemark v2.1 | 68.2 | #24 of 42 | Official leaderboard | |
| Creative Writing v3 | 62.3 | #77 of 120 | Official leaderboard | |
| Longform Creative Writing (EQ-Bench) | 48.3 | #87 of 119 | Official leaderboard | |
| Phare | 60.9% | #48 of 62 | Official leaderboard | |
| Vectara Hallucination Leaderboard (2025 dataset) | 14.5% | #92 of 104 | Official leaderboard | |
| Extended NYT Connections | 7.50% | #95 of 97 | Official leaderboard | |
| BullshitBench V2 | 2.00% | #122 of 122 | Official leaderboard | |
| BullshitBench | 3.64% | #107 of 107 | Official leaderboard |
Math and science
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| AMC | 52% | #2 of 4 | Model card | |
| DTBench | 41.8% | #123 of 178 | Official leaderboard | |
| GPQA Diamond | 43.9% | #246 of 284 | Model card | |
| Verifying the Verifiers (Project APE CRED benchmark) | 32% | #20 of 23 | Official leaderboard | |
| CritPt | 0% | #146 of 160 | Epoch AI | |
| SimpleBench | 20.4% | #92 of 100 | Official leaderboard |
Coding
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| LiveCodeBench | 34.4% | #56 of 73 | Model card | |
| IOI (Vals) | 4.00% | #45 of 55 | Official leaderboard | |
| Code Arena | 1,233 | #116 of 124 | Official leaderboard | |
| WebDev Arena | 1,230 | #99 of 105 | Epoch AI | |
| ALE-Bench | 265 | #115 of 121 | Epoch AI | |
| Image-to-WebDev Arena | 1,361 | #53 of 55 | Official leaderboard |
Agents and tools
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Elimination Game | 4 | #33 of 56 | Official leaderboard | |
| PinchBench | 65.9% | #31 of 48 | Official leaderboard | |
| Step Race | 1 | #41 of 59 | Official leaderboard | |
| Pencil Puzzle Bench | 0% | #41 of 53 | Official leaderboard | |
| PinchBench 2.0 | 54.5% | #46 of 57 | Official leaderboard | |
| PACT | 1,449 | #24 of 26 | Official leaderboard |
Professional work
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| CaseLaw v2 | 61.4% | #21 of 53 | Official leaderboard | |
| TaxEval v2 | 73.1% | #52 of 129 | Official leaderboard | |
| CorpFin v2 | 61% | #61 of 119 | Official leaderboard | |
| Vals Finance Agent v1 | 23.5% | #31 of 47 | Official leaderboard | |
| LegalBench | 79.1% | #93 of 138 | Official leaderboard | |
| MortgageTax | 52.1% | #74 of 86 | Official leaderboard | |
| Vals Finance Agent | 18.1% | #45 of 49 | Official leaderboard | |
| RadLE 2.0 | 181.0 | #12 of 12 | Official leaderboard |
Vision, audio and robotics
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Vision Arena | 1,199 | #80 of 136 | Official leaderboard |
Safety and security
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Human Pathogen Capabilities Test v2 (HPCT v2) | 30.8% | #64 of 71 | Official leaderboard | |
| DistressBench | 58.8% | #40 of 42 | Official leaderboard | |
| KORA | 10.8% | #41 of 42 | Official leaderboard |
Other
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| ACCoRD | 48.7 | #113 of 150 | Official leaderboard | |
| Conceptual Reasoning Index | 28.1 | #118 of 149 | Official leaderboard | |
| LMCA | 16.7% | #114 of 139 | Epoch AI | |
| SAGE (Vals) | 24.6% | #80 of 81 | Official leaderboard |
Its launch post
The benchmarks the lab chose to headline, with the scores it gave.