Mixtral 8x7B
by Mistral AI · mistral family
47B
parameters
text-generation code-generation reasoning multilingual math creative-writing summarization
Mixtral 8x7B is Mistral AI's mixture-of-experts model, utilizing eight expert networks of 7B parameters each with a routing mechanism that activates two experts per token. This architecture gives it 47B total parameters but only uses about 13B during inference, providing excellent efficiency. The model delivers performance competitive with much larger dense models while maintaining faster inference speeds. It excels at reasoning, multilingual tasks, and code generation, and is particularly well-suited for users who need high-quality output with reasonable hardware requirements.
Quick Start with Ollama
ollama run 8x7b-instruct-v0.1-q4_K_M | Creator | Mistral AI |
| Parameters | 47B |
| Architecture | transformer-decoder |
| Context | 32K tokens |
| Released | Dec 11, 2023 |
| License | Apache 2.0 |
| Ollama | mixtral |
Quantization Options
| Format | File Size | VRAM Required | Quality | Ollama Tag |
|---|---|---|---|---|
| Q4_K_M rec | 22.6 GB | 29.7 GB | | 8x7b-instruct-v0.1-q4_K_M |
| Q5_K_M | 26.3 GB | 34.4 GB | | 8x7b-instruct-v0.1-q5_K_M |
| Q8_0 | 44.1 GB | 49 GB | | 8x7b-instruct-v0.1-q8_0 |
Compatible Hardware
Q4_K_M requires 29.7 GB VRAM
Benchmark Scores
70.6
mmlu