OpenAI120BQ4_K_M 下约 71 GB 显存

GPT-OSS 120B — 显存、速度与本地部署

作者: Jakub Rusinowski · 最后更新:

OpenAI's first large open-weight release. Dense 120B model at Q4_K_M fits in ~65 GB VRAM — dual RTX 3090 or single A100 80GB. Matches GPT-4o on MMLU (87.4%), MATH (78.9%), and HumanEval (88.5%). The go-to choice for enterprise teams that need GPT-4-class quality without API costs or data privacy concerns.

GPT-OSS 120B 在 Q4_K_M 下约需 71 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91
创意89
编程89

按量化级别的显存与速度

计算基准:NVIDIA RTX 4090 (24 GB)。包含 8K 上下文的 KV 缓存,因此数值高于上方的主数字。

量化显存速度(估算)适配
Q2_K
2.63 bpw
39.8 GB~31 tok/s需卸载
Q3_K_M
3.41 bpw
51.2 GB~27 tok/s需卸载
Q4_K_M
4.83 bpw
71.9 GB—放不下
Q5_K_M
5.67 bpw
84.2 GB—放不下
Q6_K
6.56 bpw
97.2 GB—放不下
Q8_0
8.50 bpw
125.5 GB—放不下
F16
16.00 bpw
235 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. GPT-OSS 120B 显存计算器 →

运行 GPT-OSS 120B

目录中能运行 GPT-OSS 120B 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 GPT-OSS 120B

安装 Ollama,然后运行:

ollama run gpt-oss:120b
Hugging Face 上的权重: openai/gpt-oss-120b ↗

规格

Preview — The model is released, but these specs are thin or rest on a single source. Individual fields may be wrong.

Preview — The model is released, but these specs are thin or rest on a single source. Individual fields may be wrong.

参数量
116.8B (5.1B active)
上下文窗口
131,072
架构
Mixture-of-Experts (128 experts, top-4, MXFP4)
提供商
OpenAI
许可证
Apache 2.0
规格量化
Q4_K_M
系统内存
128 GB
记录更新于
2026-04-01
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enterpriseprivacy sensitiveon premisegeneral purpose
基准分数来源
MMLU87.4 / 100 %已发布
MATH78.9 / 100 %已发布
HumanEval88.5 / 100 %已发布

我的 GPU 能运行 GPT-OSS 120B 吗?

GPT-OSS 的其他尺寸

GPT-OSS 120B — 常见问题

How much VRAM does GPT-OSS 120B need?

About 71 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 GPT-OSS 120B run on an RTX 4090 (24 GB)?

No. GPT-OSS 120B needs about 71 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 GPT-OSS 120B locally?

Install Ollama and run `ollama run gpt-oss:120b`. That pulls the weights and starts a local OpenAI-compatible endpoint; after the download nothing leaves the machine.

What other sizes does GPT-OSS come in?

GPT-OSS 120B (71 GB), GPT-OSS 20B (13 GB). Every size shares the family's training and licence; the larger ones score higher and need proportionally more memory.