QwQ-32B Large Language Model Overview and Performance

Alibaba released the QwQ-32B model, a 32‑billion‑parameter (320 billion) LLM that rivals the 671‑billion‑parameter DeepSeek‑R1.

The notation “b” stands for billions of parameters: 1.5b = 150 M, 7b = 700 M, 32b = 3.2 B, 671b = 67.1 B, etc.

Benchmark results show QwQ-32B matches or exceeds DeepSeek‑R1 on several tests:

Math (AIME24) – comparable to DeepSeek‑R1, far better than o1‑mini and distilled 32B models.

Code (LiveCodeBench) – on par with DeepSeek‑R1, well ahead of o1‑mini.

General ability – top scores on LiveBench, IFEval, and BFCL, surpassing DeepSeek‑R1.

Performance charts compare QwQ-32B with DeepSeek‑R1‑Distilled‑Qwen‑32B, DeepSeek‑R1‑Distilled‑Llama‑70B, o1‑mini and the original DeepSeek‑R1.

Useful resources:

Blog: https://qwenlm.github.io/blog/qwq-32b

HuggingFace: https://huggingface.co/Qwen/QwQ-32B

ModelScope: https://modelscope.cn/models/Qwen/QwQ-32B

Demo: https://huggingface.co/spaces/Qwen/QwQ-32B-Demo

Qwen Chat: https://chat.qwen.ai

Example Python code for calling the model via the DashScope API:

from openai import OpenAI
import os
# Initialize OpenAI client
client = OpenAI(
    api_key=os.getenv("DASHSCOPE_API_KEY"),
    base_url="https://dashscope.aliyuncs.com/compatible-mode/v1"
)
completion = client.chat.completions.create(
    model="qwq-32b",
    messages=[{"role":"user","content":"Which is larger, 9.9 or 9.11?"}],
    stream=True
)
for chunk in completion:
    if not chunk.choices:
        print("\nUsage:", chunk.usage)
    else:
        delta = chunk.choices[0].delta
        if hasattr(delta, 'reasoning_content') and delta.reasoning_content is not None:
            print(delta.reasoning_content, end='', flush=True)
        else:
            print(delta.content, end='', flush=True)