MiMo-V2.6 Distill 9B — VRAM, speed & local setup
Written by Jakub Rusinowski · Last updated
Xiaomi's MiMo-V2.6-Distill-Qwen-9B. NOT a MiMo-V2.6 model scaled down: this is Qwen3.5-9B fine-tuned by Xiaomi on data generated by MiMo-V2.6, for coding and tool use. Dense, 8.95B in the language model plus a vision tower, 32 layers with 8 full-attention layers, 262,144-token context, MIT. It is the member of the MiMo-V2.6 release that runs on a laptop: about 6 GB at Q4_K_M. Xiaomi reports SWE-bench Verified 61.1 against 60.0 for the Qwen3.5-9B it started from.
MiMo-V2.6 Distill 9B needs about 6 GB of VRAM at Q4_K_M — quantized weights plus framework overhead, before any KV cache. On Apple Silicon that figure comes out of unified memory.
VRAM and speed by quantization
Quoted against NVIDIA RTX 4090 (24 GB). Includes the KV cache at 8K context, so it reads higher than the headline figure.
| Quant | Memory | VRAM | Speed (est.) | Fit |
|---|---|---|---|---|
| Q2_K 2.63 bpw | 4 GB | ~158 tok/s | Fits | |
| Q3_K_M 3.41 bpw | 4.9 GB | ~134 tok/s | Fits | |
| Q4_K_M 4.83 bpw | 6.5 GB | ~104 tok/s | Fits | |
| Q5_K_M 5.67 bpw | 7.4 GB | ~92 tok/s | Fits | |
| Q6_K 6.56 bpw | 8.4 GB | ~82 tok/s | Fits | |
| Q8_0 8.50 bpw | 10.6 GB | ~67 tok/s | Fits | |
| F16 16.00 bpw | 19 GB | ~38 tok/s | Fits |
Black marker = usable memory on the NVIDIA RTX 4090 (24 GB). Estimates from the memory-bandwidth roofline on the methodology page. MiMo-V2.6 Distill 9B VRAM calculator →
Get MiMo-V2.6 Distill 9B running
The cheapest catalogued GPU that runs MiMo-V2.6 Distill 9B is the Intel Arc B570 (10 GB).
or compare on Vast.ai from $0.35/hr (typical low · varies)
As an Amazon Associate we earn from qualifying purchases. Cloud GPU links are referral links — we may earn a commission at no extra cost to you.
How to run MiMo-V2.6 Distill 9B
No Ollama library entry. GGUF builds from ggml-org and others run in llama.cpp and LM Studio.
Specifications
Verified — Checked against the primary source — the model card or the vendor spec page — and corroborated by a second independent source.
- Parameters
- 8.95 Billion (+ vision tower)
- Context window
- 262,144
- Architecture
- Qwen3.5-9B hybrid attention (distilled), vision via projector
- Provider
- Xiaomi
- Licence
- MIT
- Availability
- Weights downloadable
- Practical deployment
- Laptop
- Specified at
- Q4_K_M
- Vision projector
- +0.6 GB when you send images (not included above)
- System RAM
- 16 GB
- Record updated
- 2026-10-15
Commercial use permitted. No usage restrictions beyond attribution.
Quality and use cases
Scores as published by the model’s authors or an independent evaluator — quality, not throughput, and not measured by us.
| Benchmark | Score | Provenance |
|---|---|---|
| SWE-bench Verified (avg@3) | 61.1 % | vendor-claimed · Xiaomi model card |
Other MiMo-V2.6 sizes
MiMo-V2.6 Distill 9B — frequently asked questions
How much VRAM does MiMo-V2.6 Distill 9B need?
About 6 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 Distill 9B run on an RTX 4090 (24 GB)?
Yes. MiMo-V2.6 Distill 9B needs about 6 GB at Q4_K_M, inside a 24 GB card, at an estimated 104 tokens/sec.
How do I run MiMo-V2.6 Distill 9B locally?
No Ollama library entry. GGUF builds from ggml-org and others run in llama.cpp and LM Studio. 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.