Model comparison
GPT-4 vs Mistral Large
Mistral Large is the stronger model overall, scoring 31.9 to 29.1 on the Noometry Index.
Last verified . 29 shared benchmarks.
Mistral AISummary
- They share 29 benchmarks with published results for both. GPT-4 scores higher in 2 categories and Mistral Large in 6 categories; 6 gaps are clear of the uncertainty.
- The widest gap is in knowledge, where Mistral Large leads 30.1 to 18.4.
- The biggest single-benchmark swing is MATH Level 5: 23% for GPT-4 and 50.3% for Mistral Large.
- Mistral Large is cheaper at $2 / $6 per million input/output tokens, against $30 / $60 for GPT-4.
- Mistral Large accepts more context: 131K tokens versus 8K.
- Mistral Large has downloadable open weights; the other is API-only.
Side by side
| GPT-4 | Mistral Large | |
|---|---|---|
| Provider | OpenAI | Mistral AI |
| Noometry Index | 29.1 | 31.9 |
| Released | 2023-03-14 | 2024-02-26 |
| Weights | Proprietary | Open |
| Context window | 8K | 131K |
| Max output | 8K | 16K |
| Input $ / M tokens | $30 | $2 |
| Output $ / M tokens | $60 | $6 |
| Results tracked | 38 | 51 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding Mistral Large leads
GPT-4: 31.6 (#283), Mistral Large: 34.3 (#240)
| Benchmark | GPT-4 | Mistral Large |
|---|---|---|
| BigCodeBench Instruct | 46% | 30% |
| LMArena Coding | 1254 | 1277 |
| BigCodeBench Complete | 57.2% | 38.3% |
| HumanEval+ | 79.3% | 62.2% |
| SciCode | — | 36.2% |
| WeirdML | 12.4% | — |
| LiveBench Coding | — | 47.1% |
| ALE-Bench | — | 264.7 |
| MBPP+ | — | 59.5% |
Agentic & Tool Use Not comparable
GPT-4: —, Mistral Large: 28.6 (#89)
| Benchmark | GPT-4 | Mistral Large |
|---|---|---|
| Berkeley Function Calling Leaderboard | — | 38.4% |
| METR Time Horizons | 36.1% | — |
Reasoning GPT-4 leads
GPT-4: 17.8 (#289), Mistral Large: 15.8 (#310)
| Benchmark | GPT-4 | Mistral Large |
|---|---|---|
| LMArena Hard Prompts | 1241 | 1257 |
| DTBench | 62.7% | 65.1% |
| LMCA | 17.1% | 16.7% |
| Epoch Capabilities Index | 125.89 | 128.52 |
| ForecastBench | 57.8 | 57.1 |
| SimpleBench | — | 22.5% |
| CritPt | — | 0% |
| Chess Puzzles | 4% | — |
| LiveBench Reasoning | — | 43.5% |
| Mystery Game Puzzles | 12% | — |
| LiveBench Data Analysis | — | 50.1% |
| BIG-Bench Hard | 75.1% | — |
| HellaSwag | 95.3% | — |
| LiveBench | — | 48.4% |
| WinoGrande | 87.5% | — |
Math Mistral Large leads
GPT-4: 10.8 (#309), Mistral Large: 18.2 (#291)
| Benchmark | GPT-4 | Mistral Large |
|---|---|---|
| OTIS Mock AIME 2024-2025 | 1.1% | 8.5% |
| LMArena Math | 1269 | 1262 |
| MATH Level 5 | 23% | 50.3% |
| Omni-MATH | — | 28.1% |
| LiveBench Math | — | 42.5% |
| FrontierMath (Feb 2025 set) | — | 0.3% |
| GSM8K | 92% | — |
Knowledge Mistral Large leads
GPT-4: 18.4 (#282), Mistral Large: 30.1 (#230)
| Benchmark | GPT-4 | Mistral Large |
|---|---|---|
| GPQA Diamond | 35.7% | 51.3% |
| LMArena Expert | 1211 | 1232 |
| MMLU | 86.4% | 80% |
| MMLU-Pro | — | 59.9% |
| Confabulations | — | 21.4% |
| Vectara Hallucination Rate | — | 4.5% |
| GPQA (HELM) | — | 43.5% |
| TriviaQA | 84.8% | — |
Multilingual Too close to call
GPT-4: 40.6 (#215), Mistral Large: 40.0 (#219)
| Benchmark | GPT-4 | Mistral Large |
|---|---|---|
| LMArena Non-English | 1246 | 1237 |
| LMArena Chinese | 1242 | 1240 |
| LMArena French | 1283 | 1325 |
| LMArena German | 1251 | 1254 |
| LMArena Japanese | 1209 | 1188 |
| LMArena Korean | 1184 | 1202 |
| LMArena Russian | 1251 | 1257 |
| LMArena Spanish | 1261 | 1268 |
Instruction Following Mistral Large leads
GPT-4: 65.3 (#222), Mistral Large: 67.9 (#191)
| Benchmark | GPT-4 | Mistral Large |
|---|---|---|
| LMArena Instruction Following | 1241 | 1249 |
| LiveBench Instruction Following | — | 67.9% |
| IFEval | — | 87.7% |
Long Context Too close to call
GPT-4: 37.7 (#212), Mistral Large: 38.3 (#199)
| Benchmark | GPT-4 | Mistral Large |
|---|---|---|
| LMArena Longer Query | 1244 | 1261 |
Writing & Preference Mistral Large leads
GPT-4: 34.9 (#268), Mistral Large: 40.7 (#242)
| Benchmark | GPT-4 | Mistral Large |
|---|---|---|
| LMArena Text | 1263 | 1266 |
| LMArena Creative Writing | 1244 | 1243 |
| EQ-Bench Creative Writing | 752 | 985 |
| LMArena Multi-Turn | 1257 | 1260 |
| Short-Story Creative Writing | — | 69% |
| WildBench | — | 80.1% |
| LiveBench Language | — | 39.4% |
Frequently asked questions
Is GPT-4 better than Mistral Large?
Mistral Large is the stronger model overall, scoring 31.9 to 29.1 on the Noometry Index.
Which is cheaper, GPT-4 or Mistral Large?
Mistral Large is cheaper. It lists at $2 per million input tokens and $6 per million output tokens; GPT-4 lists at $30 and $60.
Is GPT-4 or Mistral Large better for coding?
Mistral Large scores higher on coding benchmarks: 34.3 versus 31.6 in the Noometry coding category.
Which has the bigger context window?
Mistral Large does, with 131K tokens against 8K.
How many benchmarks do GPT-4 and Mistral Large share?
29 benchmarks have published results for both models. GPT-4 has 38 scored results on Noometry and Mistral Large has 51.