How much VRAM does Qwen3.5-REAP-212B-A17B need?

Q4_K_M is exactly 128,371,258,496 bytes (119.56 GiB) in weights. Add the KV cache, which depends on context length, plus roughly half a gigabyte of runtime overhead.

How large is Qwen3.5-REAP-212B-A17B's KV cache?

0.94 GiB at 32K context with an f16 cache, computed per layer from the model's architecture. Quantizing the cache to q8_0 roughly halves it.

Is Qwen3.5-REAP-212B-A17B a mixture-of-experts model?

Yes — 267 experts with 10 routed per token. Every expert must be resident in memory, but only the routed ones are read per token, so its memory requirement and its speed behave very differently.

OpenMOSE · text · mixture of experts

Qwen3.5-REAP-212B-A17B

OpenMOSE/Qwen3.5-REAP-212B-A17B

Qwen3.5-REAP-212B-A17B at Q4_K_M is exactly 128,371,258,496 bytes (119.56 GiB / 128.37 GB) — an effective 4.849 bits per weight, not the nominal 4. Its KV cache at 32K is 0.94 GiB.

From the file· summed from 1 file(s)From the file· KV per layer

Parameters

212B

total, not active

Architecture

qwen35moe

60 layers

Context

262,144

native (config.json)

License

apache-2.0

Shipped quantizations

● exact bytes, summed from published files

Quant	Size●	Exact bytes●	Effective bpw●	Tensors●	Publisher
IQ2_XXS	52.41 GiB	56,272,028,800	2.126	—	RobinsonLabs
IQ2_XS	58.30 GiB	62,597,842,048	2.365	—	RobinsonLabs
IQ2_M	64.76 GiB	69,538,235,520	2.627	—	RobinsonLabs
IQ3_XS	81.00 GiB	86,967,772,288	3.285	—	RobinsonLabs
IQ3_M	86.49 GiB	92,870,501,504	3.508	—	RobinsonLabs
IQ4_XS	105.39 GiB	113,164,088,448	4.275	—	RobinsonLabs
Q4_K_M	119.56 GiB	128,371,258,496	4.849	—	RobinsonLabs
Q5_K_M	139.97 GiB	150,288,397,440	5.677	—	RobinsonLabs

KV cache by context

computed per layer

Context	KV cache (f16)●	Flat formula	Overstated by	Full / windowed / recurrent
4,096	0.12 GiB	0.47 GiB	4.00×	15 / 0 / 45
8,192	0.23 GiB	0.94 GiB	4.00×	15 / 0 / 45
16,384	0.47 GiB	1.88 GiB	4.00×	15 / 0 / 45
32,768	0.94 GiB	3.75 GiB	4.00×	15 / 0 / 45
65,536	1.88 GiB	7.50 GiB	4.00×	15 / 0 / 45
131,072	3.75 GiB	15.00 GiB	4.00×	15 / 0 / 45

45 of 60 layers use linear attention, which keeps a fixed-size recurrent state instead of a per-token cache. Those layers do not grow with context at all — treating them as ordinary attention, as a flat formula does, overstates this model's cache by roughly 4.0× at long context.

Compare with

same modality, comparable size

vs Qwen3-Coder-30B-A3B-Instruct vs Qwen3.6-27B vs Qwen3.8-27B vs DeepSeek-V4-Flash vs Qwen3-30B-A3B-Thinking-2507 vs Laguna-XS-2.1 All alternatives →Will it fit on my card? →

Why other calculators give a different number

A parameters × bits ÷ 8 estimate puts Q4_K_M at roughly 110.94 GiB. The real file is 119.56 GiB, because a quantization is a mixture and some tensors are always kept at higher precision.

Architecture

from config.json

Layers

Attention heads

KV heads

Head dim

256

Hidden size

4096

Vocab

248,320

Sliding window

none

SWA period

—

MLA

Experts

267

Experts per token

use_sliding_window

—

Questions people ask

How much VRAM does Qwen3.5-REAP-212B-A17B need?: Q4_K_M is exactly 128,371,258,496 bytes (119.56 GiB) in weights. Add the KV cache, which depends on your context length, plus roughly half a gigabyte of runtime overhead.
How large is Qwen3.5-REAP-212B-A17B's KV cache?: 0.94 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 Qwen3.5-REAP-212B-A17B a mixture-of-experts model?: Yes — 267 experts, 10 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 Qwen3.5-REAP-212B-A17B 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.