Mistral Small 4 — Mistral AI 的本地 AI 模型
作者： Jakub Rusinowski · 最后更新：
Mistral AI's March 16, 2026 release unifying its former Magistral/Pixtral/Devstral lines into a single 119B-total / ~6.5B-active MoE model (128 experts, 4 active per token) with text + image input, reasoning, and agentic coding. Apache 2.0 licensed, 256K context. Every expert must be resident, so at Q4_K_M it is a 128 GB-class model (a high-memory Mac or several GPUs), not a single consumer graphics card.
变体
Mistral Small 4 最小的变体在 Q4_K_M 下约需 73 GB 显存——量化权重加框架开销，不含 KV 缓存。
| 模型 | Q4 下显存 | 显存 | 上下文 | 运行 |
|---|---|---|---|---|
| Mistral Small 4 119B-A6.5B → 119B (~6.5B active) | ~72.6 GB | 256,000 | 云端托管标签 |
显存为 Q4_K_M 下的量化权重加开销，与 GPU 与显存检测器使用同一引擎计算。
如何在本地运行 Mistral Small 4
安装 Ollama，然后拉取标签。
No Ollama library tag for Mistral Small 4 (checked 2026-11-03): the mistral-small tag is the older 22B/24B Mistral Small. Use community GGUFs with llama.cpp, or vLLM.
在上方选择一个尺寸，查看它自己的显存、速度估算和安装命令。
许可证
Commercial use permitted. No usage restrictions beyond attribution.
适用于: Mistral Small 4 119B-A6.5B推荐 GPU
目录中能在本地运行 Mistral Small 4（至少 73 GB 显存）的最便宜 GPU 是 NVIDIA DGX Spark (128 GB).
我的 GPU 能运行 Mistral Small 4 吗？
- Mistral Small 4 在 AMD Ryzen AI Max+ 395 上
- Mistral Small 4 在 Apple M1 Max 上
- Mistral Small 4 在 Apple M1 Pro 上
- Mistral Small 4 在 Apple M1 Ultra 上
- Mistral Small 4 在 Apple M2 Max 上
- Mistral Small 4 在 Apple M2 Pro 上
- Mistral Small 4 在 Apple M3 Max 上
- Mistral Small 4 在 Apple M3 Pro 上
- Mistral Small 4 在 Apple M4 上
- Mistral Small 4 在 Apple M4 Max 上
- Mistral Small 4 在 Apple M5 上
- Mistral Small 4 在 Apple M5 Max 上
- Mistral Small 4 在 Apple M5 Pro 上
- Mistral Small 4 在 NVIDIA A100 80GB (PCIe) 上
- Mistral Small 4 在 NVIDIA DGX Spark 上
- Mistral Small 4 在 NVIDIA GeForce RTX 5090 上
- Mistral Small 4 在 NVIDIA H100 80GB (PCIe) 上
- Mistral Small 4 在 NVIDIA L40S 上
- Mistral Small 4 在 NVIDIA RTX 6000 Ada Generation 上
- Mistral Small 4 在 NVIDIA RTX PRO 6000 Blackwell 上
Mistral Small 4 — 常见问题
How much VRAM does Mistral Small 4 need?
Mistral Small 4 needs about 73 GB VRAM at Q4_K_M quantization for its smallest variant. Variants: Mistral Small 4 119B-A6.5B (73 GB, Q4_K_M). On Apple Silicon, unified memory counts toward this requirement.
Can I run Mistral Small 4 on an RTX 4090 (24 GB)?
Mistral Small 4's smallest variant needs about 73 GB, which exceeds a single RTX 4090 (24 GB). Use multiple GPUs, a higher-VRAM card, or Apple Silicon with large unified memory.
What quantization should I use for Mistral Small 4?
Q4_K_M is the best balance of quality and VRAM for Mistral Small 4 in most cases. Choose Q8_0 for near-lossless quality if you have spare VRAM, or smaller quants (Q3/Q2) only when memory is tight.
How do I run Mistral Small 4 with Ollama?
Mistral Small 4 has no local Ollama tag — the published tag is cloud-hosted, so running it sends your prompts to a hosted GPU rather than your own machine.