QwQ

What QwQ is for

QwQ is a reasoning model: it spends output tokens working through a problem before answering, which makes it slower per useful word and better on problems that need the working shown. Alibaba released it in 2025-03 with a single 32B size under Apache 2.0, with a 32K context window. Budget for the extra tokens the thinking phase consumes. The catalogue records its strength as QwQ reasoning model on a single 24GB GPU.

Sizes and memory

The 32B needs about 20 GB of weights at Q4_K_M. That is the whole range; there is no larger or smaller variant in this record. Recommended VRAM adds roughly half again on top of the weights for the KV cache and the runtime.

Size	Weights at Q4_K_M	Recommended VRAM	Smallest card in our guide that holds it
32B	20 GB	30 GB	NVIDIA RTX 3090

Hardware that runs QwQ

11 of the 19 cards in our hardware guide hold at least one size of QwQ at Q4_K_M. The cheapest that does is the NVIDIA RTX 3090 at about $700, running the 32B.

The largest size any of them holds is the 32B at 20 GB, on the NVIDIA RTX 3090 (24 GB at 936 GB/s). Bandwidth, not capacity, sets the speed once a size fits, so a card further down the list with faster memory will generate more tokens per second on the same weights than a larger, slower one.

What people use it for

The catalogue lists 3 recorded uses for QwQ: reasoning, math, logic. It is tagged reasoning, thinking, math.

Where QwQ sits in the QwQ family

QwQ is the only QwQ entry in our catalogue, so there is no in-family alternative to weigh it against; the comparison to make is against the models listed at the foot of this page.

The Apache 2.0 licence puts no usage ceiling on it, which matters if the thing you are building has users. That is the practical difference between this and a vendor community licence.

Context window: 32K

32K tokens is about 24,000 words at the usual ratio — enough for a handful of files or a long report held whole. It is the point at which the KV cache starts to be a serious share of memory: doubling the context you actually use costs roughly double the cache, on top of the 20 GB of weights.

How to run QwQ locally

Our catalogue does not record registry tags, so look the current tag up in the Ollama library or on Hugging Face before pasting these. The commands below are the shape of the workflow, not a copy-and-paste recipe.

Option 1: Ollama (simplest)

# find the tag for QwQ at ollama.com/library
ollama run <tag>

Option 2: Mullama (production)

mullama pull <tag>
mullama run <tag>

Option 3: llama.cpp (CLI)

# download a GGUF from Hugging Face, searching for "QwQ GGUF"
./llama-cli -m model.Q4_K_M.gguf -p "Hello, AI!"

Option 4: Python with Mullama or llama-cpp-python

from mullama import Model, Context
model = Model.load("model.Q4_K_M.gguf", n_gpu_layers=99)
ctx = Context(model, n_ctx=4096)
print(ctx.generate("Hello, AI!", 256))

Sources

Model record from src/data/models.json (2026-06-29); card capacities from src/data/gpus.json (2026-06-29). Upstream: Curated from ollama.com/library + community benchmarks + paperswithcode + huggingface.