Mistral Large 4.
Mistral AI Oct 2026 Flagship
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.
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| BullshitBench V2 | 80% | #11 of 129 | Official leaderboard | |
| LMArena (Elo) | 1,423 | #101 of 335 | Official leaderboard | |
| DesignArena ASCII Art | 1,204 | #28 of 77 | Official leaderboard | |
| Last Translation Benchmark | 5.71 | #21 of 36 | Official leaderboard | |
| Lech Mazur Creative Writing | -1.68 | #37 of 50 | Official leaderboard | |
| Extended NYT Connections | 27.4% | #84 of 103 | Official leaderboard | |
| SimpleQA Verified | 22.4% | #78 of 94 | Epoch AI run | |
| HANDBOOK.md | 19.1% | #15 of 16 | Official leaderboard | |
| ComplexConstraints | 42.4% | #15 of 15 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| OTIS Mock AIME 2024-2025 | 99.4% | #21 of 212 | Epoch AI run | |
| GPQA Diamond | 90.2% | #48 of 229 | Epoch AI run | |
| DTBench | 88.2% | #58 of 180 | Official leaderboard | |
| scBench (LatchBio SingleCellBench) | 53% | #18 of 26 | Official leaderboard | |
| Prophet Arena | 85.4 | #17 of 24 | Official leaderboard | |
| LatchBio Capabilities v1.0 | 34.9% | #22 of 27 | Official leaderboard | |
| FrontierPhysics | 0% | #22 of 22 | Official leaderboard | |
| Riemann-bench | 20% | #20 of 20 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Vibe Code Bench | 81.6% | #29 of 120 | Official leaderboard | |
| DesignArena UI Component | 1,279 | #35 of 139 | Official leaderboard | |
| DesignArena Data Visualization | 1,274 | #30 of 118 | Official leaderboard | |
| DesignArena SVG | 1,224 | #29 of 103 | Official leaderboard | |
| DesignArena 3D Design | 1,242 | #40 of 136 | Official leaderboard | |
| DesignArena Website | 1,259 | #48 of 151 | Official leaderboard | |
| Code Arena | 1,564 | #42 of 128 | Official leaderboard | |
| DesignArena Code | 1,255 | #49 of 143 | Official leaderboard | |
| ALE-Bench | 1,657 | #49 of 126 | Official leaderboard | |
| DesignArena Game Dev | 1,205 | #61 of 143 | Official leaderboard | |
| APEX-SWE | 37% | #41 of 65 | Official leaderboard | |
| DeepSWE | 61.7% | #37 of 56 | Lab model card | |
| sudo L7 | 13.9% | #13 of 18 | Official leaderboard | |
| FrontierCode 2.0 | 19.6% | #24 of 32 | Official leaderboard | |
| SWE Atlas v1.1 | 45.2% | #23 of 28 | Official leaderboard | |
| DAXBench v2 | 63.2% | #40 of 47 | Official leaderboard | |
| DesignArena Image to HTML | 1,114 | #47 of 51 | Official leaderboard | |
| WeirdML v3 | 4.20% | #26 of 27 | Official leaderboard | |
| DAYJOB: Data Science | 0.70% | #15 of 15 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| AutomationBench | 59.9% | #4 of 47 | Lab model card | |
| Terminal-Bench 4.0 | 27.9% | #19 of 40 | Official leaderboard | |
| Terminal-Bench 4.1 | 27.6% | #21 of 43 | Official leaderboard | |
| Chess Puzzles | 29% | #54 of 93 | Epoch AI run | |
| Mystery Game Puzzles | 20% | #50 of 84 | Epoch AI run | |
| EBR-bench | 18.1% | #21 of 31 | Epoch AI run | |
| DuelLab GameBench 2 | 57.6 | #41 of 57 | Official leaderboard | |
| Agent Arena | -5.44% | #43 of 56 | Official leaderboard | |
| Collaborative Intelligence Benchmark | 44.5% | #14 of 18 | Official leaderboard | |
| Clocktower Radio | 1,265 | #21 of 23 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Harvey LAB v1.1 (HLAB-Vals-1.1) | 15.8% | #10 of 38 | Official leaderboard | |
| Legora BAR | 103.0 | #5 of 17 | Official leaderboard | |
| Tax Agent Bench | 30.7% | #33 of 84 | Official leaderboard | |
| EnterpriseBench: CoreCraft | 47.7% | #25 of 51 | Official leaderboard | |
| Vals Finance Agent v2.1 | 61.7% | #20 of 40 | Official leaderboard | |
| Legal Research Bench | 35.6% | #50 of 93 | Official leaderboard | |
| DAYJOB: Healthcare | 3.60% | #21 of 37 | Official leaderboard | |
| Vals Index v2.2 | 56.5% | #21 of 37 | Official leaderboard | |
| Vals Finance Index | 57% | #23 of 40 | Official leaderboard | |
| APEX-Agents 1.1 | 64.3% | #38 of 65 | Official leaderboard | |
| EEBench | 50.1% | #19 of 30 | Official leaderboard | |
| DAYJOB: Finance | 4.80% | #25 of 39 | Official leaderboard | |
| Excel Modeling Benchmark v1.1 | 68.7% | #27 of 42 | Official leaderboard | |
| Harvey LAB (Legal Agent Benchmark) | 15.8% | #9 of 14 | Lab model card | |
| APEX-Accounting | 56.5% | #32 of 49 | Official leaderboard | |
| Tuesday Work Index | 43.4 | #29 of 44 | Official leaderboard | |
| Vals Legal Index | 67.8% | #28 of 38 | Official leaderboard | |
| APEX-1 | 59.6% | #55 of 71 | Official leaderboard | |
| Vals Finance Agent v2 | 54.7% | #9 of 11 | Lab model card | |
| DeepFinance | 21.5% | #19 of 23 | Official leaderboard | |
| TaxCalcBench v2 | 6.00% | #27 of 32 | Official leaderboard | |
| DAYJOB: Law | 4.50% | #15 of 16 | Official leaderboard | |
| DAYJOB: Accounting | 3.60% | #17 of 17 | Official leaderboard | |
| DAYJOB: Tax | 0% | #15 of 15 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Vision Arena | 1,231 | #61 of 136 | Official leaderboard | |
| Furniture Assembly Benchmark | 30.8% | #29 of 39 | Epoch AI run | |
| GDP.pdf | 32.8% | #18 of 19 | Official leaderboard | |
| Dense 200 (bbox) | 42% | #2 of 2 | Lab model card |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| BioTIER Permit | 99.9% | #10 of 69 | Official leaderboard | |
| ABC Bench Fragment Design | 98.5% | #4 of 27 | Official leaderboard | |
| DistressBench | 77% | #9 of 42 | Official leaderboard | |
| KORA | 60.7% | #10 of 42 | Official leaderboard | |
| OncologyBench | 22.7% | #9 of 30 | Official leaderboard | |
| ABC Bench Liquid-Handling Robot | 98% | #12 of 34 | Official leaderboard | |
| ABC Bench Screening Evasion | 66.7% | #6 of 17 | Official leaderboard | |
| Human Pathogen Capabilities Test | 56% | #33 of 86 | Official leaderboard | |
| SecIT Bench | 76.4% | #10 of 26 | Official leaderboard | |
| Molecular Biology Capabilities Test | 51.3% | #35 of 90 | Official leaderboard | |
| Agentic BAIM-LLM Evaluation (ABLE) | 65.3% | #7 of 18 | Official leaderboard | |
| FORTRESS | 23.1% | #36 of 73 | Official leaderboard | |
| World-Class Bio | 44.3% | #40 of 74 | Official leaderboard | |
| BioTIER Refuse | 32.2% | #45 of 69 | Official leaderboard | |
| VCT-v2 | 39.4% | #18 of 25 | Official leaderboard | |
| EpiBench | 20.8% | #20 of 27 | Official leaderboard | |
| BioSecBench-Function | 28.8% | #21 of 27 | Official leaderboard | |
| CWE-bench 2 | 45.6% | #11 of 14 | Official leaderboard | |
| TxBench-Preclinical-Pharmacology | 42.3% | #24 of 30 | Official leaderboard | |
| VariantBench | 28% | #22 of 27 | Official leaderboard | |
| TxBench-Antibody-Discovery | 28.3% | #25 of 30 | Official leaderboard | |
| MetagenomicsBench | 26.3% | #23 of 26 | Official leaderboard | |
| SpatialBench-Long | 20.8% | #24 of 27 | Official leaderboard | |
| TxBench-Oligo-Discovery | 27.1% | #27 of 30 | Official leaderboard | |
| BioSecBench-Refusal | 3.40% | #20 of 22 | Official leaderboard | |
| BioSecBench-Surveillance | 34% | #25 of 27 | Official leaderboard | |
| scBench-Long | 10.2% | #25 of 27 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Conceptual Reasoning Index | 50.7 | #67 of 151 | Official leaderboard | |
| ACCoRD | 65.7 | #72 of 152 | Official leaderboard | |
| LMCA | 35.8% | #77 of 147 | Official leaderboard | |
| LiveBench | 40.9% | #30 of 31 | Official leaderboard |
The Self-Reported Index fits one score to the 5 capability results Mistral AI reported for this model, on the scale of the Epoch Capabilities Index. Labs choose which benchmarks to show and how to run them, so read it as the lab's claim, not an independent measurement. The 90% interval runs from 151.9 to 164.8. How we compute it.
Where the benchmark's own leaderboard, or Epoch AI's run, also has this model. Gaps mostly come from different harnesses, effort levels and task subsets. Every pair.
| Benchmark | Lab says | Board says | Gap |
|---|---|---|---|
| Terminal-Bench 4.0 | 28.3% Coding - Terminal Bench 4.0: AA - Terminal-Bench 4 (chart 28; text 28.3%) | 27.9% Leaderboard, max, Claude Code | +0.4 pts |
Mistral AI's models in color, against every other model on the Epoch Capabilities Index, by release date.
Source: Epoch Capabilities Index, Epoch AI (CC BY 4.0)
The benchmarks the lab chose to headline, with the scores it gave.