ModelVRAM
EN 中文

2026 开源大模型消费级显卡显存报告：64 个开源模型在 8–80 GB 显卡和 Mac 上

本报告的每个数字都在构建时由站内 64 个开源模型的数据和显存计算器的同一套函数算出，没有手填数字。设定：Q4_K_M 权重（gpt-oss 按发布的 MXFP4），32K 上下文，单个请求，FP16 KV 缓存，显存至少剩 0.5 GB 才算放得下。数据版本 1.2026-09-30，模型数据核对于 。GB = GiB。

可引用的结论

  1. 在 32K 上下文下，63 个开源模型单个请求的 FP16 KV 缓存从 0.19 GB（Nemotron 3 Nano 30B-A3B）到 11.0 GB（Mistral Medium 3.5 128B）不等，相差 58.7 倍。
  2. 参数量决定不了 KV 缓存：Nemotron 3 Nano 30B-A3B（31.6B 参数）在 32K 下只需 0.19 GB KV 缓存，而 Granite 4.2 30B（29.3B）需要 8.0 GB，是前者的 43 倍。
  3. 64 个开源模型中有 34 个能以 Q4_K_M、32K 上下文放进 24 GB 显卡，其中最大的是 Ornith 1.5 35B-A3B，需要 23.5 GB。
  4. 在 Q4_K_M、32K 上下文下，48 GB 显卡能放下的最大模型并不比 32 GB 更大：两者都是 K2-Horizon MoVA 36B-A4B，多出的显存只换来更长的上下文（37K 到 115K token）；Llama 3.1 70B 需要 55.2 GB。
  5. 128 GB 内存的 Mac（GPU 可用 96 GB）能以 Q4_K_M、32K 上下文运行 64 个模型中的 44 个，最大到 Mistral Medium 3.5 128B（127.7B，91.8 GB）。

如何引用

纯文本：数据来源：ModelVRAM（modelvram.com），《2026 开源大模型消费级显卡显存报告》，https://modelvram.com/reports/local-llm-vram-2026/，数据版本 1.2026-09-30（2026-09-30），CC BY 4.0。

APA: ModelVRAM. (2026). 2026 Local LLM VRAM Report (Version 1.2026-09-30) [Data set]. https://modelvram.com/reports/local-llm-vram-2026/

数据以 CC BY 4.0 许可发布：可以复制、改编和商用，只需署名“Data: ModelVRAM”并链接到本页。

各档显存能放下的最大模型（Q4_K_M，32K）

按总参数排序取最大的一个；“最长上下文”是该模型在这档显存上还剩 0.5 GB 时能开到的上下文。Mac 按 macOS 允许 GPU 使用的统一内存计算。

显存放得下的模型数 最大的模型设定 显存（32K）最长上下文
8 GB 显卡 11 / 64 LensVLM 9B Q4_K_M 7.43 GB 33K
12 GB 显卡 18 / 64 Gemma 4 12B Q4_K_M 8.98 GB 178K
16 GB 显卡 19 / 64 gpt-oss-20b MXFP4 15.44 GB 34K
24 GB 显卡 34 / 64 Ornith 1.5 35B-A3B (MoE) Q4_K_M 23.47 GB 33K
32 GB 显卡 37 / 64 K2-Horizon MoVA 36B-A4B (MoE) Q4_K_M 30.31 GB 37K
48 GB 显卡 37 / 64 K2-Horizon MoVA 36B-A4B (MoE)与更小一档相同 Q4_K_M 30.31 GB 115K
80 GB 显卡 43 / 64 Qwen3.5 122B-A10B (MoE) Q4_K_M 78.85 GB 57K
64 GB Mac（可用 48 GB） 37 / 64 K2-Horizon MoVA 36B-A4B (MoE)与更小一档相同 Q4_K_M 30.31 GB 115K
128 GB Mac（可用 96 GB） 44 / 64 Mistral Medium 3.5 128B Q4_K_M 91.75 GB 41K

48 GB 这一档的空缺

48 GB 显卡（和 64 GB Mac）能放下的最大模型与 32 GB 相同，都是 K2-Horizon MoVA 36B-A4B (MoE)，多出的显存只让上下文从 37K 增加到 115K。 原因是数据集里 40B–75B 参数的模型只有 1 个（Llama 3.1 70B 在 Q4_K_M、32K 下需要 55.2 GB），都放不进 48 GB。 这是本数据集的覆盖空缺，不代表 48 GB 显卡跑不了这个尺寸的其他模型；我们没有为这一档补编数据。

32K 上下文下的 KV 缓存：从小到大

单个请求、FP16。最小的 Nemotron 3 Nano 30B-A3B (MoE) 为 0.19 GB，最大的 Mistral Medium 3.5 128B 为 11.0 GB，相差 58.7 倍。差距来自注意力结构：只有部分层保存完整 KV 的混合模型，缓存远小于每层都是全注意力的模型。

#模型参数KV 缓存（32K）结构（英文）
1 Nemotron 3 Nano 30B-A3B (MoE) 31.6B3.5B 激活 0.19 GB 6 of its 52 layers use full attention and 46 are Mamba or feed-forward layers with no growing cache
2 Ling 3.0 Tiny 7.9B-A1.3B (MoE) 7.9B1.3B 激活 0.21 GB 6 of its 24 layers use multi-head latent attention and 18 are KDA recurrent layers with no growing cache
3 Nemotron 3 Super 120B-A12B (MoE) 123.6B12.0B 激活 0.25 GB 8 of its 88 layers use full attention and 80 are Mamba or feed-forward layers with no growing cache
4 LFM2.5 8B-A1B (MoE) 8.5B1.5B 激活 0.38 GB 6 of its 24 layers use full attention and 18 are short-convolution layers with no growing cache
5 GLM-5.3 Flash 321.3B18.0B 激活 0.39 GB 11 of its 45 layers use multi-head latent attention and 34 are linear-attention layers with no growing cache
6 Limite 1B Violetto 1.0B 0.44 GB 12 of its 48 layers use full attention and 36 keep a sliding window of 1,025 tokens
7 Nemotron 3 Nano 4B 4.0B 0.50 GB 4 of its 42 layers use full attention and 38 are Mamba or feed-forward layers with no growing cache
8 Muse Glimmer 30B 29.8B 0.50 GB 13 of its 52 layers use full attention and 39 keep a sliding window of 2,048 tokens
9 Gemma 4 E4B 8.0B 0.54 GB 4 of its 42 layers use full attention, 20 keep a sliding window of 512 tokens and 18 reuse the cache of earlier layers
10 Ornith 1.0 35B (MoE) 35.1B3.0B 激活 0.63 GB 10 of its 40 layers use full attention and 30 are linear-attention layers with no growing cache
11 Ornith 1.5 35B-A3B (MoE) 36.0B3.0B 激活 0.63 GB 10 of its 40 layers use full attention and 30 are linear-attention layers with no growing cache
12 Qwen3.6 35B-A3B (MoE) 36.0B3.0B 激活 0.63 GB 10 of its 40 layers use full attention and 30 are linear-attention layers with no growing cache
13 AliceAI Foundation 80B-A3B (MoE) 81.3B3.0B 激活 0.75 GB 12 of its 48 layers use full attention and 36 are KDA recurrent layers with no growing cache
14 Qwen3-Coder-Next (80B MoE) 79.7B3.0B 激活 0.75 GB 12 of its 48 layers use full attention and 36 are linear-attention layers with no growing cache
15 Qwen3.5 122B-A10B (MoE) 125.1B10.0B 激活 0.75 GB 12 of its 48 layers use full attention and 36 are linear-attention layers with no growing cache
16 Qwen3.8 Flash Next (180B MoE) 180.0B6.0B 激活 0.75 GB 12 of its 48 layers use full attention and 36 are linear-attention layers with no growing cache
17 gpt-oss-20b 20.9B3.6B 激活 0.77 GB 12 of its 24 layers use full attention and 12 keep a sliding window of 128 tokens
18 Kimi K3 2.78T104.0B 激活 0.84 GB 24 of its 93 layers use multi-head latent attention and 69 are linear-attention layers with no growing cache
19 MiMo V2.6 Flash 310.8B15.0B 激活 0.85 GB 9 of its 48 layers use full attention and 39 keep a sliding window of 128 tokens
20 Gemma 4 26B-A4B (MoE) 25.8B4.0B 激活 0.92 GB 5 of its 30 layers use full attention and 25 keep a sliding window of 1,024 tokens
21 Gemma 4 12B 12.0B 0.97 GB 8 of its 48 layers use full attention and 40 keep a sliding window of 1,024 tokens
22 LensVLM 9B 9.4B 1.00 GB 8 of its 32 layers use full attention and 24 are linear-attention layers with no growing cache
23 MiMo V2.6 Distill Qwen 9B 9.4B 1.00 GB 8 of its 32 layers use full attention and 24 are linear-attention layers with no growing cache
24 Ornith 1.0 9B 9.4B 1.00 GB 8 of its 32 layers use full attention and 24 are linear-attention layers with no growing cache
25 Ornith 1.5 9B 9.7B 1.00 GB 8 of its 32 layers use full attention and 24 are linear-attention layers with no growing cache
26 Qwen3.5 9B 9.7B 1.00 GB 8 of its 32 layers use full attention and 24 are linear-attention layers with no growing cache
27 ZDTaichu 5.0 9B 9.8B 1.00 GB 8 of its 32 layers use full attention and 24 are linear-attention layers with no growing cache
28 gpt-oss-120b 116.8B5.1B 激活 1.15 GB 18 of its 36 layers use full attention and 18 keep a sliding window of 128 tokens
29 Spark-X2.5 4B 4.1B 1.23 GB 9 of its 36 layers use full attention and 27 keep a sliding window of 512 tokens
30 MiniCPM5 2B 2.5B 1.31 GB All 42 layers use full attention
31 Xing 4.0 29B-A4B (MoE) 31.2B4.0B 激活 1.41 GB All 40 layers use multi-head latent attention
32 Step 3.7 Flash 196B-A11B (MoE) 201.4B11.0B 激活 1.63 GB 12 of its 45 layers use full attention and 33 keep a sliding window of 512 tokens
33 GLM-4.7 Flash 31.2B3.0B 激活 1.65 GB All 47 layers use multi-head latent attention
34 MiMo V2.6 Pro 1.02T42.0B 激活 1.78 GB 10 of its 70 layers use full attention and 60 keep a sliding window of 128 tokens
35 Hemmingway-1 27B 27.3B 2.00 GB 16 of its 64 layers use full attention and 48 are linear-attention layers with no growing cache
36 LLM-jp-4.1 32B-A3B Thinking (MoE) 32.1B3.8B 激活 2.00 GB All 32 layers use full attention
37 Qwen3.6 27B 27.8B 2.00 GB 16 of its 64 layers use full attention and 48 are linear-attention layers with no growing cache
38 Qwen3.8 27B 27.8B 2.00 GB 16 of its 64 layers use full attention and 48 are linear-attention layers with no growing cache
39 ThinkingCap Qwen3.8 27B 27.8B 2.00 GB 16 of its 64 layers use full attention and 48 are linear-attention layers with no growing cache
40 DeepSeek V3 / R1 (671B) 684.5B37.0B 激活 2.14 GB All 61 layers use multi-head latent attention
41 DeepSeek V3.2 685.4B37.0B 激活 2.38 GB All 61 layers use multi-head latent attention
42 DeepSeek V4.1 Flash 763.2B16.0B 激活 2.50 GB All 40 layers use full attention
43 Granite 4.2 3B 3.7B 2.50 GB All 40 layers use full attention
44 DeepSeek V4 Flash 290.9B13.0B 激活 2.69 GB All 43 layers use full attention
45 DeepSeek V4 Flash 0731 304.2B13.0B 激活 2.69 GB All 43 layers use full attention
46 GLM-5.2 753.3B40.0B 激活 2.82 GB All 78 layers use multi-head latent attention
47 GLM-5.3 753.3B40.0B 激活 2.82 GB All 78 layers use multi-head latent attention
48 Hy4 Preview 770B-A49B (MoE) 780.0B49.0B 激活 2.82 GB All 78 layers use multi-head latent attention
49 Qwen3.8 2.4T-A95B (MoE) 2.45T95.0B 激活 2.88 GB 23 of its 92 layers use full attention and 69 are linear-attention layers with no growing cache
50 Qwen3 30B-A3B (MoE) 30.5B3.3B 激活 3.00 GB All 48 layers use full attention
51 Qwen3-Coder 30B-A3B (MoE) 30.5B3.3B 激活 3.00 GB All 48 layers use full attention
52 Gemma 4 31B 31.3B 3.67 GB 10 of its 60 layers use full attention and 50 keep a sliding window of 1,024 tokens
53 MiniMax M3 427.0B23.0B 激活 3.75 GB All 60 layers use full attention
54 DeepSeek V4 Pro 1.60T49.0B 激活 3.81 GB All 61 layers use full attention
55 Llama 3.1 8B 8.0B 4.00 GB All 32 layers use full attention
56 IQuest-Q1 320B-A15B (MoE) 320.3B15.0B 激活 4.23 GB 25 of its 88 layers use full attention and 63 keep a sliding window of 4,096 tokens
57 Qwen3 8B 8.2B 4.50 GB All 36 layers use full attention
58 Granite 4.2 8B 8.8B 5.00 GB All 40 layers use full attention
59 K2-Horizon MoVA 36B-A4B (MoE) 37.4B4.0B 激活 6.00 GB All 48 layers use full attention
60 MiniMax M2.7 228.7B10.0B 激活 7.75 GB All 62 layers use full attention
61 Granite 4.2 30B 29.3B 8.00 GB All 64 layers use full attention
62 Llama 3.1 70B 70.6B 10.00 GB All 80 layers use full attention
63 Mistral Medium 3.5 128B 127.7B 11.00 GB All 88 layers use full attention

下载与嵌入

把图表放进你的文章，复制这段代码（图片链接回本报告，并附署名）：

<a href="https://modelvram.com/reports/local-llm-vram-2026/"><img src="https://modelvram.com/reports/local-llm-vram-2026/kv-cache-32k.svg" width="760" height="1226" alt="KV cache at 32K context for 63 open LLMs, from 0.19 GB to 11.0 GB" loading="lazy"></a>
<p>Source: <a href="https://modelvram.com/reports/local-llm-vram-2026/">ModelVRAM 2026 Local LLM VRAM Report</a> (CC BY 4.0)</p>

63 个开源模型在 32K 上下文下的 KV 缓存，从 0.19 GB 到 11.0 GB

方法与局限

模型的层数、注意力结构和参数量读取自 Hugging Face 上各仓库的 config.json 与文件大小。总显存 = 权重 + FP16 KV 缓存 + 0.5 GB + 权重与缓存的 10%。KV 缓存按实际结构计算：全注意力层随上下文增长，滑动窗口层只保留窗口，MLA 保存压缩的潜向量，线性注意力和状态空间层不随上下文增长。

局限：只算单个请求；不含 vLLM 的 gpu_memory_utilization 预留、CUDA Graph 和视觉编码器；这些是估算值，不是实测峰值。与 20 条公开 llama.cpp / vLLM 实测的对比见估算准确度。任意其他设置可以在显存计算器里重算。

更新：本页与开放数据集同步，数据变化时随站点重新生成；引用时请注明上面的数据版本号，以便对应到具体数字。首次发布于 。