mistralai · text · mixture of experts

Mixtral-8x22B-Instruct-v0.1

mistralai/Mixtral-8x22B-Instruct-v0.1

Mixtral-8x22B-Instruct-v0.1 at Q4_K_M is exactly 85,593,307,840 bytes (79.71 GiB / 85.59 GB) — an effective 4.869 bits per weight, not the nominal 4. Its KV cache at 32K is 7.00 GiB.

From the file· summed from 2 file(s)From the file· KV per layer
Parameters
141B
total, not active
Architecture
llama
56 layers
Context
65,536
native (config.json)
License
apache-2.0

Shipped quantizations

● exact bytes, summed from published files
QuantSize●Exact bytes●Effective bpw●Tensors●Publisher
IQ1_S27.61 GiB29,648,894,4321.687—MaziyarPanahi
IQ1_M30.49 GiB32,737,212,8961.862—MaziyarPanahi
Q2_K3 shards48.53 GiB52,108,606,2722.964—MaziyarPanahi
IQ3_XS3 shards54.23 GiB58,234,125,1203.313—MaziyarPanahi
Q3_K_S3 shards57.28 GiB61,504,109,3763.499—MaziyarPanahi
Q3_K_M2 shards63.14 GiB67,795,565,2483.857—MaziyarPanahi
Q3_K_L2 shards67.60 GiB72,585,722,5604.129—MaziyarPanahi
IQ4_XS2 shards71.12 GiB76,360,596,1604.344—MaziyarPanahi
Q4_K_S2 shards74.96 GiB80,484,645,5684.579—MaziyarPanahi
Q4_K_M2 shards79.71 GiB85,593,307,8404.869—MaziyarPanahi
Q5_K_S4 shards90.32 GiB96,980,843,4245.517—MaziyarPanahi
Q5_K_M4 shards93.11 GiB99,975,576,4805.687—MaziyarPanahi
Q8_04 shards139.16 GiB149,424,847,7768.500—MaziyarPanahi

KV cache by context

computed per layer
ContextKV cache (f16)●Flat formulaOverstated byFull / windowed / recurrent
4,0960.88 GiB0.88 GiB—56 / 0 / 0
8,1921.75 GiB1.75 GiB—56 / 0 / 0
16,3843.50 GiB3.50 GiB—56 / 0 / 0
32,7687.00 GiB7.00 GiB—56 / 0 / 0
65,53614.00 GiB14.00 GiB—56 / 0 / 0
131,07228.00 GiB28.00 GiB—56 / 0 / 0

Compare with

same modality, comparable size
vs Qwen3-Coder-30B-A3B-Instructvs Qwen3.6-27Bvs Qwen3.8-27Bvs gemma-4-E4B-itvs Qwen3-30B-A3B-Thinking-2507vs Qwen3-8BAll alternatives →Will it fit on my card? →

Will it run on your card?

full quant x context sweep
By VRAM:8GB12GB16GB24GB32GB48GB
Radeon RX 6500 XT 4GBGeForce RTX 3050 6GBGeForce RTX 5050 8GBGeForce RTX 3080 10GBGeForce RTX 2080 Ti 11GBGeForce RTX 5070 12GBGeForce RTX 5060 Ti 16GBApple M3 Pro 18GBGeForce RTX 3080 Ti 20GBGeForce RTX 5090 D V2 24GBGeForce RTX 5090 D 32GBApple M5 Max 36GBApple M5 Pro 48GBApple M5 Max 64GBApple M3 Ultra 96GBApple M5 Max 128GBApple M2 Ultra 192GBApple M3 Ultra 256GBApple M3 Ultra 512GB

Why other calculators give a different number

A parameters × bits ÷ 8 estimate puts Q4_K_M at roughly 73.67 GiB. The real file is 79.71 GiB, because a quantization is a mixture and some tensors are always kept at higher precision.

Architecture

from config.json
Layers
56
Attention heads
48
KV heads
8
Head dim
128
Hidden size
6144
Vocab
32,768
Sliding window
none
SWA period
—
MLA
no
Experts
8
Experts per token
2
use_sliding_window
—

Questions people ask

How much VRAM does Mixtral-8x22B-Instruct-v0.1 need?
Q4_K_M is exactly 85,593,307,840 bytes (79.71 GiB) in weights. Add the KV cache, which depends on your context length, plus roughly half a gigabyte of runtime overhead.
How large is Mixtral-8x22B-Instruct-v0.1's KV cache?
7.00 GiB at 32K context with an f16 cache, computed per layer. Quantizing the cache to q8_0 roughly halves it, which is often the difference between a context length fitting and not.
Is Mixtral-8x22B-Instruct-v0.1 a mixture-of-experts model?
Yes — 8 experts, 2 routed per token. Every expert must be resident, but only the routed ones are read per token, which is why its memory requirement and its speed behave very differently.
Which quantization of Mixtral-8x22B-Instruct-v0.1 should I use?
Q4_K_M is the usual default. Pick the largest quantization that fits your card at the context you actually need — the table above gives exact sizes for every one published.