MiMo V2.6 Flash VRAM requirements
MiMo V2.6 Flash needs about 193 GB of VRAM at Q4_K_M with 8K context, so it needs a 640 GB setup such as 8× H100 SXM; a 24 GB card is too small. MiMo V2.6 Flash has 310.8B parameters (MiMo-V2.6-Flash-MOPD has the same config and parameter count, and its checkpoint and GGUF files are the same size, so these figures cover it too), of which about 15B are used per token; all 256 experts still have to be in memory. With the same 8K context and one request it needs 319 GB at FP8 and 638 GB at FP16/BF16. The published weights take 161 GB (4.5 bits/weight). It does not fit on a 24 GB card at Q4_K_M; the smallest setup here that holds it with an 8K context is the M3 Ultra Mac Studio (512 GB, 384 GB usable).
193 GB at Q4_K_M, 8K context, one request
- FP8
- 319 GB
- FP16 / BF16
- 638 GB
- Published weights
- 161 GB
- Smallest setup, Q4_K_M
- M3 Ultra Mac Studio (512 GB, 384 GB usable)
Worked example: MiMo V2.6 Flash with 32K tokens
Inputs: MiMo V2.6 Flash · Q4_K_M weights · 32,768 tokens of context · one request · FP16 KV cache
- Weights (310.8B at Q4_K_M)
- 175 GB
- KV cache (23 KB per token)
- 866 MB
- Buffers and runtime (0.5 GB + 10%)
- 18.1 GB
- Total
- 194 GB
The smallest setup here that holds it is the M3 Ultra Mac Studio (512 GB, 384 GB usable), with about 190 GB to spare. Change the inputs in the calculator
What makes MiMo V2.6 Flash's memory use different
Its 39 sliding-window layers keep only the last 128 tokens (llama.cpp gives them 768 cells: the window plus a 512-token batch, rounded up to 256): at its 1M limit they hold 146 MB of cache instead of the 195 GB they would need with full attention.
As a mixture-of-experts model it reads about 8.45 GB of its 175 GB Q4_K_M weights per generated token (5%), so it writes like a much smaller model while needing memory for all of them.
How much VRAM does MiMo V2.6 Flash need?
MiMo V2.6 Flash needs 193 GB at Q4_K_M, 339 GB at Q8_0 and 638 GB at FP16 with 8,192 tokens of context and one request. Each total below is the weights plus the FP16 KV cache and the runtime overhead (0.5 GB plus 10%), and each row opens the calculator with that setting. What the GGUF names mean.
| Precision | Weights | Total | Smallest setup |
|---|---|---|---|
| As published (4.5 bits/weight) | 161 GB | 178 GB | B200 (180 GB) |
| FP16 / BF16 | 579 GB | 638 GB | 8× H200 (1,128 GB) |
| FP8 / INT8 | 289 GB | 319 GB | M3 Ultra Mac Studio (512 GB, 384 GB usable) |
| INT4 (AWQ / GPTQ) | 154 GB | 170 GB | B200 (180 GB) |
| GGUF Q8_0 | 308 GB | 339 GB | M3 Ultra Mac Studio (512 GB, 384 GB usable) |
| GGUF Q6_K | 237 GB | 262 GB | M3 Ultra Mac Studio (512 GB, 384 GB usable) |
| GGUF Q5_K_M | 205 GB | 226 GB | M3 Ultra Mac Studio (512 GB, 384 GB usable) |
| GGUF Q4_K_M | 175 GB | 193 GB | M3 Ultra Mac Studio (512 GB, 384 GB usable) |
| GGUF IQ4_XS | 157 GB | 174 GB | B200 (180 GB) |
| GGUF Q3_K_M | 141 GB | 156 GB | B200 (180 GB) |
| GGUF IQ3_XXS | 119 GB | 132 GB | H200 (141 GB) |
| GGUF Q2_K | 121 GB | 134 GB | H200 (141 GB) |
Fine-tuning MiMo V2.6 Flash? MiMo V2.6 Flash VRAM for LoRA, QLoRA and full training.
MiMo V2.6 Flash GGUF files on Hugging Face
The 1 GGUF files llama.cpp picks for MiMo V2.6 Flash with -hf <repo>:<QUANT>, with the bytes the Hugging Face file list reported on 2026-09-29. Each total is that file plus the FP16 KV cache and 0.5 GB + 10%, one request. "vs estimate" compares the file with the bits-per-weight size the table above uses.
| Quant | File | Size | vs estimate | 8K | 32K | 128K |
|---|---|---|---|---|---|---|
| Q2_K | ggml-org/ | 118 GB126,214,593,856 bytes | −3% | 130 GB | 131 GB | 133 GB |
GB here is GiB (1024³ bytes), as everywhere on this site.
Which MiMo V2.6 Flash GGUF fits a 16, 24, 32, 48 or 80 GB GPU or a Mac
From the file sizes above: the largest file that fits each machine with an 8K context and 0.5 GB left free, the longest context it then has room for, and the longest context for Q4_K_M. One request, FP16 KV cache, -np 1.
| Hardware | Largest GGUF, 8K | Its longest context | Q4_K_M, longest context |
|---|---|---|---|
| RTX 5060 Ti 16GB | None fits | — | — |
| RTX 4090 (24 GB) | None fits | — | — |
| RTX 5090 (32 GB) | None fits | — | — |
| L40S (48 GB) | None fits | — | — |
| H100 SXM (80 GB) | None fits | — | — |
| M4 Pro Mac (64 GB, 48 GB usable) | None fits | — | — |
| M4 Max Mac (128 GB, 96 GB usable) | None fits | — | — |
MiMo V2.6 Flash at 8K, 32K, 128K and 1M (full) tokens of context
9 of its 48 layers use full attention and 39 keep a sliding window of 128 tokens. Each extra token of context adds 23 KB of FP16 cache per request once the sliding windows are full. The last column is the smallest setup here, one card or a group, that holds it at Q4_K_M. How the KV cache works.
| Context | KV cache, FP16 | Total at Q4_K_M | Total at Q8_0 | Smallest setup, Q4_K_M |
|---|---|---|---|---|
| 8K tokens | 326 MB | 193 GB | 339 GB | M3 Ultra Mac Studio (512 GB, 384 GB usable) |
| 32K tokens | 866 MB | 194 GB | 340 GB | M3 Ultra Mac Studio (512 GB, 384 GB usable) |
| 128K tokens | 2.96 GB | 196 GB | 342 GB | M3 Ultra Mac Studio (512 GB, 384 GB usable) |
| 1M tokens | 22.6 GB | 218 GB | 364 GB | M3 Ultra Mac Studio (512 GB, 384 GB usable) |
How fast MiMo V2.6 Flash writes
Tokens per second for one request with 8,192 tokens of context, estimated from memory bandwidth and the parameters read per token. A dash means it does not fit on one card. Try other settings in the speed calculator.
| Hardware | Bandwidth | Q4_K_M | FP8 |
|---|---|---|---|
| RTX 3060 12GB | 360 GB/s | — | — |
| RTX 4090 | 1,008 GB/s | — | — |
| RTX 5090 | 1,792 GB/s | — | — |
| M4 Max Mac (128 GB) | 546 GB/s | — | — |
| M3 Ultra Mac Studio (512 GB) | 819 GB/s | 25–43 | 16–26 |
| H100 SXM | 3,350 GB/s | — | — |
| H200 | 4,800 GB/s | — | — |
Longest context on a multi-GPU group
No single card in this table holds it at Q4_K_M, Q8_0 or FP8, so these rows use GPU groups: how many tokens of context MiMo V2.6 Flash fits on each, in one tensor-parallel group with one request, an FP16 KV cache and 0.5 GB left free. "Full" means the model's whole context window fits.
| Group | Memory | Q4_K_M | Q8_0 | FP8 |
|---|---|---|---|---|
| 2× RTX 3060 12GB | 24 GB | No | No | No |
| 2× RTX 3090 | 48 GB | No | No | No |
| 2× RTX 4090 | 48 GB | No | No | No |
| 2× RTX 5090 | 64 GB | No | No | No |
| 4× RTX 3090 | 96 GB | No | No | No |
| 8× H100 SXM | 640 GB | 1M (full) | 1M (full) | 1M (full) |
| 8× H200 | 1,128 GB | 1M (full) | 1M (full) | 1M (full) |
Model details
- Parameters
- 310.8B (310,756,322,688); MiMo-V2.6-Flash-MOPD has the same config and parameter count, and its checkpoint and GGUF files are the same size, so these figures cover it too
- Experts
- 256 routed experts, all loaded
- Active per token
- 15B
- Layers
- 9 of its 48 layers use full attention and 39 keep a sliding window of 128 tokens
- Attention cache
- 4 KV heads × 192 keys, 128 values
- Sliding-window layers
- 8 KV heads × 192
- Context length
- 1,048,576 tokens
- Published weights
- 161 GB (4.5 bits/weight)
- On Hugging Face
- XiaomiMiMo/
MiMo-V2.6-Flash-RL - Reported benchmarks
- None published in the model card for GPQA, MMLU-Pro or LiveCodeBench.
Why this estimate looks this way
At Q4_K_M and an 8K-token context, MiMo V2.6 Flash uses 175 GB for weights, 326 MB for its FP16 KV cache and 18.0 GB for estimated runtime overhead, totaling 193 GB. The overhead is a 0.5 GB base plus 10% of weights and cache.
This is a mixture-of-experts model: all 256 experts contribute to the 310.8B parameters held in memory, even though only about 15B parameters run per token. Using only active parameters would understate VRAM.
Its attention layout matters for long context: 9 of its 48 layers use full attention and 39 keep a sliding window of 128 tokens. The FP16 cache grows by about 23 KB per additional token per request once the sliding windows are full.
The architecture and published checkpoint size come from the model's config.json and weight files. These are estimates rather than measured peak memory; inference engines can reserve extra buffers or preallocate the full configured cache.
MiMo V2.6 Flash next to similar-sized models
The three models here closest to it in parameter count, at Q4_K_M with 8,192 tokens of context.
| Model | Parameters | Total | KV per token | Smallest setup |
|---|---|---|---|---|
| MiMo V2.6 Flash | 310.8B, 15B active | 193 GB | 23 KB | M3 Ultra Mac Studio (512 GB, 384 GB usable) |
| DeepSeek V4 Flash 0731 | 304.2B, 13B active | 190 GB | 86 KB | M3 Ultra Mac Studio (512 GB, 384 GB usable) |
| IQuest-Q1 320B-A15B | 320.3B, 15B active | 201 GB | 100 KB | M3 Ultra Mac Studio (512 GB, 384 GB usable) |
| GLM-5.3 Flash | 321.3B, 18B active | 200 GB | 19 KB | M3 Ultra Mac Studio (512 GB, 384 GB usable) |
MiMo V2.6 Flash VRAM questions
How much VRAM do I need to run MiMo V2.6 Flash locally?
About 193 GB at Q4_K_M with an 8K context and one request, by the bits-per-weight estimate, and 638 GB at BF16. The smallest setup listed here that holds Q4_K_M is the M3 Ultra Mac Studio (512 GB, 384 GB usable).
Can I run MiMo V2.6 Flash on a 24 GB GPU?
No. The RTX 4090 (24 GB) alone cannot hold it: the smallest listed GGUF, Q2_K at 118 GB, needs 130 GB with an 8K context.
Which quantization of MiMo V2.6 Flash fits in 16 GB of VRAM?
The RTX 5060 Ti 16GB alone cannot hold it: the smallest listed GGUF, Q2_K at 118 GB, needs 130 GB with an 8K context.
Can I run MiMo V2.6 Flash on a Mac?
The M4 Pro Mac (64 GB, 48 GB usable) alone cannot hold it: the smallest listed GGUF, Q2_K at 118 GB, needs 130 GB with an 8K context. The M4 Max Mac (128 GB, 96 GB usable) alone cannot hold it: the smallest listed GGUF, Q2_K at 118 GB, needs 130 GB with an 8K context. By default macOS lets the GPU use about 75% of unified memory.
How much more VRAM does MiMo V2.6 Flash need for 32K or 128K tokens of context?
Its FP16 KV cache is 326 MB at 8K, 866 MB at 32K, 2.96 GB at 128K for one request, so 128K adds 2.64 GB over 8K, plus 10% overhead. A q8_0 cache
Does MiMo V2.6 Flash need less VRAM because only 15B parameters are active?
No. All 256 experts stay loaded, so the Q4_K_M weights take 175 GB. The active parameters decide speed: each generated token reads about 8.45 GB of them.
Badge for your model card
Paste it into a Hugging Face or GitHub README; it shows the Q4_K_M total above and links to this page. Q8_0 and FP16 badges.
Numbers read from the model files on Hugging Face on . Compare it with other models in the reproducible model and GPU report.