What is Qwen 2.5?
Qwen 2.5 — Alibaba's reasoning and multilingual model
Qwen 2.5 is an open-source model family from Alibaba Cloud, released in late 2024. It tops many open-source benchmarks in mathematics, coding and reasoning, and has the strongest multilingual performance of any commonly-available Ollama model — supporting over 29 languages including Chinese, Japanese, Arabic and European languages.
On Windows with Ollama, Qwen 2.5 7B downloads at about 4.4 GB and runs comfortably on a GPU with 6 GB VRAM. The larger 14B variant delivers noticeably better quality and fits in 10 GB VRAM.
Quick start
Run Qwen 2.5 on Windows
# Pull Qwen 2.5 7B (~4.4 GB):
C:\> ollama pull qwen2.5
success
# Or the coding variant:
C:\> ollama pull qwen2.5-coder
# Start chat:
C:\> ollama run qwen2.5
>>> Solve this math problem: ...
Variants
Qwen 2.5 variants and requirements
| Variant | Size | Min VRAM | Pull command |
|---|---|---|---|
| Qwen 2.5 0.5B | 0.4 GB | 2 GB | ollama pull qwen2.5:0.5b |
| Qwen 2.5 1.5B | 1.0 GB | 2 GB | ollama pull qwen2.5:1.5b |
| Qwen 2.5 7B (default) | 4.4 GB | 6 GB | ollama pull qwen2.5 |
| Qwen 2.5 14B | 8.7 GB | 10 GB | ollama pull qwen2.5:14b |
| Qwen 2.5 Coder 7B | 4.4 GB | 6 GB | ollama pull qwen2.5-coder |
| Qwen 2.5 72B | 43 GB | 48 GB | ollama pull qwen2.5:72b |
Qwen 2.5 7B is the best starting point. If you need stronger reasoning or math, step up to 14B — it fits in most gaming GPUs with 10+ GB VRAM.
Use cases
What Qwen 2.5 is best at
- Mathematics and reasoning — top open-source performance on MATH and GSM8K benchmarks
- Coding — Qwen 2.5 Coder is a dedicated variant that rivals GPT-4 on coding benchmarks for its size
- Multilingual tasks — best open-source support for Chinese, Japanese, Arabic and other non-English languages
- Long context — 128K token context window for processing long documents
FAQ
Qwen 2.5 questions
Qwen 2.5 vs Llama 3 — which is better?
Qwen 2.5 outperforms Llama 3 8B on mathematics, coding and multilingual tasks. Llama 3 8B has better general conversational quality and broader English training data. For technical work or non-English languages, Qwen 2.5 is the stronger choice.
Is Qwen 2.5 safe to use? What is the license?
Qwen 2.5 models under 72B use a permissive commercial license allowing free use. The 72B variant has an additional commercial use clause for very large deployments. Check the full terms at qwenlm.github.io.
How do I use Qwen 2.5 for Chinese text?
Just type in Chinese directly:
ollama run qwen2.5 then type your Chinese prompt. Qwen 2.5 handles Chinese natively without any special configuration — it was trained with extensive Chinese data.