Model comparison
GPT-4 vs Mixtral 8x22B
GPT-4 is the stronger model overall, scoring 28.9 to 26.7 on the Noometry Index. Mixtral 8x22B costs 13× less per token, which makes it the better buy when GPT-4's lead doesn't matter for your workload.
Last verified . 27 shared benchmarks.
Mistral AISummary
- They share 27 benchmarks with published results for both. GPT-4 scores higher in 5 categories and Mixtral 8x22B in 3 categories; 8 gaps are clear of the uncertainty.
- The widest gap is in math, where Mixtral 8x22B leads 21.2 to 10.1.
- The biggest single-benchmark swing is WeirdML: 12.4% for GPT-4 and 3.2% for Mixtral 8x22B.
- Mixtral 8x22B is cheaper at $2 / $6 per million input/output tokens, against $30 / $60 for GPT-4.
- Mixtral 8x22B accepts more context: 64K tokens versus 8K.
- Mixtral 8x22B has downloadable open weights; the other is API-only.
Side by side
| GPT-4 | Mixtral 8x22B | |
|---|---|---|
| Provider | OpenAI | Mistral AI |
| Noometry Index | 28.9 | 26.7 |
| Released | 2023-03-14 | 2024-04-17 |
| Weights | Proprietary | Open |
| Context window | 8K | 64K |
| Max output | 8K | 64K |
| Input $ / M tokens | $30 | $2 |
| Output $ / M tokens | $60 | $6 |
| Results tracked | 38 | 34 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding GPT-4 leads
GPT-4: 31.2 (#277), Mixtral 8x22B: 23.8 (#324)
| Benchmark | GPT-4 | Mixtral 8x22B |
|---|---|---|
| WeirdML | 12.4% | 3.2% |
| BigCodeBench Instruct | 46% | 40.6% |
| LMArena Coding | 1254 | 1166 |
| BigCodeBench Complete | 57.2% | 50.2% |
| HumanEval+ | 79.3% | 72% |
| MBPP+ | — | 64.3% |
Agentic & Tool Use Not comparable
GPT-4: —, Mixtral 8x22B: 22.8 (#119)
| Benchmark | GPT-4 | Mixtral 8x22B |
|---|---|---|
| Cybench | — | 7.5% |
| METR Time Horizons | 36.1% | — |
Reasoning Mixtral 8x22B leads
GPT-4: 17.3 (#285), Mixtral 8x22B: 19.1 (#246)
| Benchmark | GPT-4 | Mixtral 8x22B |
|---|---|---|
| LMArena Hard Prompts | 1241 | 1150 |
| DTBench | 62.7% | 55.1% |
| Epoch Capabilities Index | 125.88 | 122 |
| ForecastBench | 57.8 | 56.3 |
| Chess Puzzles | 4% | — |
| Mystery Game Puzzles | 12% | — |
| LMCA | 17.1% | — |
| BIG-Bench Hard | 75.1% | — |
| HellaSwag | 95.3% | — |
| WinoGrande | 87.5% | — |
Math Mixtral 8x22B leads
GPT-4: 10.1 (#304), Mixtral 8x22B: 21.2 (#271)
| Benchmark | GPT-4 | Mixtral 8x22B |
|---|---|---|
| LMArena Math | 1269 | 1184 |
| MATH Level 5 | 23% | 24.2% |
| OTIS Mock AIME 2024-2025 | 1.1% | — |
| Omni-MATH | — | 16.3% |
| GSM8K | 92% | — |
Knowledge GPT-4 leads
GPT-4: 18.4 (#277), Mixtral 8x22B: 15.1 (#288)
| Benchmark | GPT-4 | Mixtral 8x22B |
|---|---|---|
| GPQA Diamond | 35.7% | 34.1% |
| LMArena Expert | 1211 | 1113 |
| MMLU | 86.4% | 77.8% |
| MMLU-Pro | — | 46% |
| GPQA (HELM) | — | 33.4% |
| TriviaQA | 84.8% | — |
Multilingual GPT-4 leads
GPT-4: 40.8 (#210), Mixtral 8x22B: 32.9 (#250)
| Benchmark | GPT-4 | Mixtral 8x22B |
|---|---|---|
| LMArena Non-English | 1246 | 1128 |
| LMArena Chinese | 1242 | 1116 |
| LMArena French | 1282 | 1165 |
| LMArena German | 1251 | 1141 |
| LMArena Japanese | 1209 | 1037 |
| LMArena Korean | 1184 | 1056 |
| LMArena Russian | 1251 | 1158 |
| LMArena Spanish | 1261 | 1150 |
Instruction Following GPT-4 leads
GPT-4: 65.5 (#217), Mixtral 8x22B: 57.9 (#261)
| Benchmark | GPT-4 | Mixtral 8x22B |
|---|---|---|
| LMArena Instruction Following | 1241 | 1147 |
| IFEval | — | 72.4% |
Long Context GPT-4 leads
GPT-4: 37.7 (#207), Mixtral 8x22B: 34.6 (#242)
| Benchmark | GPT-4 | Mixtral 8x22B |
|---|---|---|
| LMArena Longer Query | 1244 | 1144 |
Writing & Preference Mixtral 8x22B leads
GPT-4: 35.0 (#263), Mixtral 8x22B: 37.1 (#257)
| Benchmark | GPT-4 | Mixtral 8x22B |
|---|---|---|
| LMArena Text | 1263 | 1162 |
| LMArena Creative Writing | 1244 | 1141 |
| LMArena Multi-Turn | 1257 | 1130 |
| EQ-Bench Creative Writing | 752 | — |
| WildBench | — | 71.1% |
Frequently asked questions
Is GPT-4 better than Mixtral 8x22B?
GPT-4 is the stronger model overall, scoring 28.9 to 26.7 on the Noometry Index. Mixtral 8x22B costs 13× 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 8x22B?
Mixtral 8x22B 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 Mixtral 8x22B better for coding?
GPT-4 scores higher on coding benchmarks: 31.2 versus 23.8 in the Noometry coding category.
Which has the bigger context window?
Mixtral 8x22B does, with 64K tokens against 8K.
How many benchmarks do GPT-4 and Mixtral 8x22B share?
27 benchmarks have published results for both models. GPT-4 has 38 scored results on Noometry and Mixtral 8x22B has 34.