Mistral AI119B (~6.5B active)Q4_K_M 下约 73 GB 显存

Mistral Small 4 119B-A6.5B — 显存、速度与本地部署

作者: Jakub Rusinowski · 最后更新:

Unifies Mistral's reasoning, multimodal, and agentic-coding lines into one MoE checkpoint. 128 experts, 4 active per token (~6.5B active). At Q4_K_M it fits a single 24GB consumer GPU (RTX 4090/A10); FP8 and higher precision need 48GB+ workstation cards. 256K context, Apache 2.0.

Mistral Small 4 119B-A6.5B 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。在 Apple Silicon 上,这部分来自统一内存。

逻辑90
创意86
编程91

按量化级别的显存与速度

计算基准:NVIDIA RTX 4090 (24 GB)。仅含权重与开销:该模型架构未公开,因此未计入 KV 缓存。

量化显存速度(估算)适配
Q2_K
2.63 bpw
39.9 GB~11 tok/s需卸载
Q3_K_M
3.41 bpw
51.5 GB~9 tok/s需卸载
Q4_K_M
4.83 bpw
72.6 GB—放不下
Q5_K_M
5.67 bpw
85.1 GB—放不下
Q6_K
6.56 bpw
98.4 GB—放不下
Q8_0
8.50 bpw
127.2 GB—放不下
F16
16.00 bpw
238.8 GB—放不下

黑色标记 = NVIDIA RTX 4090 (24 GB) 上的可用显存。 估算来自内存带宽屋顶线模型,详见 方法说明页. Mistral Small 4 119B-A6.5B 显存计算器 →

运行 Mistral Small 4 119B-A6.5B

目录中能运行 Mistral Small 4 119B-A6.5B 的最便宜 GPU 是 AMD Ryzen AI Max+ 395 (128 GB).

联盟营销声明: 本页部分链接为联盟推广链接——如果你通过它们购买,LLM Configurator 可能会获得佣金,而你无需支付任何额外费用。作为亚马逊联盟成员(Amazon Associate),LLM Configurator 会从符合条件的购买中获得收益。
Ryzen AI Max+ 395 Laptop (Strix Halo, up to 128GB)
128 GB VRAM · 120 W board power
2026年价格波动较大——请以当前商品页价格为准。

如何运行 Mistral Small 4 119B-A6.5B

安装 Ollama,然后运行:

ollama run mistral-small
Hugging Face 上的权重: mistralai/Mistral-Small-4-119B-2603 ↗

规格

Verified — Checked against the primary source — the model card or the vendor spec page — and corroborated by a second independent source.

参数量
119 Billion (~6.5B active)
上下文窗口
256,000
架构
Mixture-of-Experts (128 experts, 4 active)
提供商
Mistral AI
许可证
Apache 2.0
规格量化
Q4_K_M
系统内存
32 GB
记录更新于
2026-03-16
许可证Apache-2.0允许商业使用

Commercial use permitted. No usage restrictions beyond attribution.

质量与使用场景

评分由模型作者或独立评测方发布——衡量质量而非吞吐量,并非我们实测。

最适合reasoningcodingmultimodalconsumer gpuagentic tasks

我的 GPU 能运行 Mistral Small 4 119B-A6.5B 吗?

Mistral Small 4 119B-A6.5B — 常见问题

How much VRAM does Mistral Small 4 119B-A6.5B need?

About 73 GB at Q4_K_M — quantized weights plus framework overhead, before any KV cache. The cache grows with context length and is added on top; the table above folds it in. Apple Silicon counts unified memory toward the same figure.

Does Mistral Small 4 119B-A6.5B run on an RTX 4090 (24 GB)?

No. Mistral Small 4 119B-A6.5B needs about 73 GB at Q4_K_M, more than a single RTX 4090's 24 GB. It needs a larger card, several GPUs, or Apple Silicon with enough unified memory — or it runs with part of the weights offloaded to system RAM, which is much slower.

How do I run Mistral Small 4 119B-A6.5B locally?

Install Ollama and run `ollama run mistral-small`. That pulls the weights and starts a local OpenAI-compatible endpoint; after the download nothing leaves the machine.