MiMo-V2.6-Flash — 显存、速度与本地部署
作者： Jakub Rusinowski · 最后更新：
The smaller MiMo-V2.6: 309B parameters (310.8B in the published tensors) with 15B active per token, 256 experts with 8 routed, 48 layers of which 39 use a 128-token sliding window and 9 use full attention. Text, image, video and audio in; 1M-token context; MIT. Every expert must be resident, about 188 GB at Q4_K_M, so it is a 256 GB unified-memory or multi-GPU model. Xiaomi publishes two checkpoints of the same size: -RL, and -MOPD, which it recommends for agents.
MiMo-V2.6-Flash 在 Q4_K_M 下约需 188 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。
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 | 显存 | 速度（估算） | 适配 |
|---|---|---|---|---|
| Q2_K 2.63 bpw | 103.2 GB | — | 放不下 | |
| Q3_K_M 3.41 bpw | 133.5 GB | — | 放不下 | |
| Q4_K_M 4.83 bpw | 188.6 GB | — | 放不下 | |
| Q5_K_M 5.67 bpw | 221.3 GB | — | 放不下 | |
| Q6_K 6.56 bpw | 255.8 GB | — | 放不下 | |
| Q8_0 8.50 bpw | 331.2 GB | — | 放不下 | |
| F16 16.00 bpw | 622.5 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. MiMo-V2.6-Flash 显存计算器 →
运行 MiMo-V2.6-Flash
目录中能运行 MiMo-V2.6-Flash 的最便宜 GPU 是 Apple M3 Ultra (512 GB).
作为亚马逊联盟成员，我们从符合条件的购买中获得收入。云 GPU 链接为推荐链接——我们可能获得佣金，您无需额外付费。
如何运行 MiMo-V2.6-Flash
No Ollama library entry. Xiaomi's documented serving routes are SGLang and vLLM; a ggml-org GGUF exists for llama.cpp.
规格
Corroborated — Two or more independent sources agree on these figures, but the model card itself was not retrieved. Treat the numbers as good rather than confirmed. 尚未确认： active parameters (15B: model card only).
- 参数量
- 309 Billion (15B active)
- 上下文窗口
- 1,048,576
- 架构
- Sparse MoE (256 experts, 8 active), hybrid sliding-window attention; vision + audio encoders
- 提供商
- Xiaomi
- 许可证
- MIT
- Availability
- Weights downloadable
- Practical deployment
- Workstation (large unified memory or pro GPU)
- 规格量化
- Q4_K_M
- 系统内存
- 256 GB
- 记录更新于
- 2026-10-15
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。
MiMo-V2.6 的其他尺寸
MiMo-V2.6-Flash — 常见问题
How much VRAM does MiMo-V2.6-Flash need?
About 188 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 MiMo-V2.6-Flash run on an RTX 4090 (24 GB)?
No. MiMo-V2.6-Flash needs about 188 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 MiMo-V2.6-Flash locally?
No Ollama library entry. Xiaomi's documented serving routes are SGLang and vLLM; a ggml-org GGUF exists for llama.cpp. Running the published tag would send your prompts to a hosted GPU rather than your own machine.
What other sizes does MiMo-V2.6 come in?
MiMo-V2.6-Pro (619 GB), MiMo-V2.6-Flash (188 GB), MiMo-V2.6 Distill 9B (6 GB). Every size shares the family's training and licence; the larger ones score higher and need proportionally more memory.