Model comparison
GPT-4 vs Mixtral 8x7B
GPT-4 is the stronger model overall, scoring 28.9 to 26.8 on the Noometry Index. Mixtral 8x7B costs 54× less per token, which makes it the better buy when GPT-4's lead doesn't matter for your workload.
Last verified . 28 shared benchmarks.
Mistral AISummary
- They share 28 benchmarks with published results for both. GPT-4 scores higher in 5 categories and Mixtral 8x7B in 3 categories; 6 gaps are clear of the uncertainty.
- The widest gap is in instruction following, where GPT-4 leads 65.5 to 51.2.
- The biggest single-benchmark swing is DTBench: 62.7% for GPT-4 and 49.6% for Mixtral 8x7B.
- Mixtral 8x7B is cheaper at $0.70 / $0.70 per million input/output tokens, against $30 / $60 for GPT-4.
- Mixtral 8x7B accepts more context: 32K tokens versus 8K.
- Mixtral 8x7B has downloadable open weights; the other is API-only.
Side by side
| GPT-4 | Mixtral 8x7B | |
|---|---|---|
| Provider | OpenAI | Mistral AI |
| Noometry Index | 28.9 | 26.8 |
| Released | 2023-03-14 | 2023-12-11 |
| Weights | Proprietary | Open |
| Context window | 8K | 32K |
| Max output | 8K | 32K |
| Input $ / M tokens | $30 | $0.70 |
| Output $ / M tokens | $60 | $0.70 |
| Results tracked | 38 | 38 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding Mixtral 8x7B leads
GPT-4: 31.2 (#277), Mixtral 8x7B: 32.2 (#265)
| Benchmark | GPT-4 | Mixtral 8x7B |
|---|---|---|
| LMArena Coding | 1254 | 1127 |
| HumanEval+ | 79.3% | 39.6% |
| WeirdML | 12.4% | — |
| BigCodeBench Instruct | 46% | — |
| BigCodeBench Complete | 57.2% | — |
| MBPP+ | — | 49.7% |
Agentic & Tool Use Not comparable
GPT-4: —, Mixtral 8x7B: —
| Benchmark | GPT-4 | Mixtral 8x7B |
|---|---|---|
| METR Time Horizons | 36.1% | — |
Reasoning Too close to call
GPT-4: 17.3 (#285), Mixtral 8x7B: 17.4 (#283)
| Benchmark | GPT-4 | Mixtral 8x7B |
|---|---|---|
| LMArena Hard Prompts | 1241 | 1115 |
| DTBench | 62.7% | 49.6% |
| Epoch Capabilities Index | 125.88 | 118.38 |
| ForecastBench | 57.8 | 56.3 |
| HellaSwag | 95.3% | 86.7% |
| WinoGrande | 87.5% | 77.2% |
| Chess Puzzles | 4% | — |
| Mystery Game Puzzles | 12% | — |
| LMCA | 17.1% | — |
| Adversarial NLI | — | 55.2% |
| BIG-Bench Hard | 75.1% | — |
| PIQA | — | 83.6% |
Math Mixtral 8x7B leads
GPT-4: 10.1 (#304), Mixtral 8x7B: 17.2 (#284)
| Benchmark | GPT-4 | Mixtral 8x7B |
|---|---|---|
| LMArena Math | 1269 | 1147 |
| MATH Level 5 | 23% | 10% |
| GSM8K | 92% | 74.4% |
| OTIS Mock AIME 2024-2025 | 1.1% | — |
| Omni-MATH | — | 10.5% |
Knowledge GPT-4 leads
GPT-4: 18.4 (#277), Mixtral 8x7B: 11.0 (#296)
| Benchmark | GPT-4 | Mixtral 8x7B |
|---|---|---|
| GPQA Diamond | 35.7% | 30.6% |
| LMArena Expert | 1211 | 1088 |
| MMLU | 86.4% | 70.6% |
| TriviaQA | 84.8% | 82.2% |
| MMLU-Pro | — | 33.5% |
| GPQA (HELM) | — | 29.6% |
| ARC (AI2) Challenge | — | 87.3% |
| OpenBookQA | — | 85.8% |
Multilingual GPT-4 leads
GPT-4: 40.8 (#210), Mixtral 8x7B: 29.8 (#261)
| Benchmark | GPT-4 | Mixtral 8x7B |
|---|---|---|
| LMArena Non-English | 1246 | 1077 |
| LMArena Chinese | 1242 | 1055 |
| LMArena French | 1282 | 1165 |
| LMArena German | 1251 | 1114 |
| LMArena Japanese | 1209 | 931 |
| LMArena Korean | 1184 | 968 |
| LMArena Russian | 1251 | 1090 |
| LMArena Spanish | 1261 | 1110 |
Instruction Following GPT-4 leads
GPT-4: 65.5 (#217), Mixtral 8x7B: 51.2 (#292)
| Benchmark | GPT-4 | Mixtral 8x7B |
|---|---|---|
| LMArena Instruction Following | 1241 | 1110 |
| IFEval | — | 57.5% |
Long Context GPT-4 leads
GPT-4: 37.7 (#207), Mixtral 8x7B: 33.4 (#255)
| Benchmark | GPT-4 | Mixtral 8x7B |
|---|---|---|
| LMArena Longer Query | 1244 | 1103 |
Writing & Preference Too close to call
GPT-4: 35.0 (#263), Mixtral 8x7B: 34.4 (#265)
| Benchmark | GPT-4 | Mixtral 8x7B |
|---|---|---|
| LMArena Text | 1263 | 1132 |
| LMArena Creative Writing | 1244 | 1109 |
| LMArena Multi-Turn | 1257 | 1115 |
| EQ-Bench Creative Writing | 752 | — |
| WildBench | — | 67.3% |
Frequently asked questions
Is GPT-4 better than Mixtral 8x7B?
GPT-4 is the stronger model overall, scoring 28.9 to 26.8 on the Noometry Index. Mixtral 8x7B costs 54× less per token, which makes it the better buy when GPT-4's lead doesn't matter for your workload.
Which is cheaper, GPT-4 or Mixtral 8x7B?
Mixtral 8x7B is cheaper. It lists at $0.70 per million input tokens and $0.70 per million output tokens; GPT-4 lists at $30 and $60.
Is GPT-4 or Mixtral 8x7B better for coding?
Mixtral 8x7B scores higher on coding benchmarks: 32.2 versus 31.2 in the Noometry coding category.
Which has the bigger context window?
Mixtral 8x7B does, with 32K tokens against 8K.
How many benchmarks do GPT-4 and Mixtral 8x7B share?
28 benchmarks have published results for both models. GPT-4 has 38 scored results on Noometry and Mixtral 8x7B has 38.