Alibaba (Qwen)
Qwen 2.5 Coder 32B
openrouter/qwen-2.5-coder-32bContext
33K
Input / 1M tokens
$0.6600
Output / 1M tokens
$1.00
Cached input / 1M tokens
— (bills at input price)
Cache write / 1M tokens
— (bills at input price)
About this model
Qwen 2.5 Coder 32B
Qwen 2.5 Coder 32B is tuned for code generation, completion, repair, and reasoning. It performs comparably to GPT-4o on coding benchmarks at a fraction of the cost.
Key facts
- Maker
- Alibaba (Qwen)
- Context window
- 33K tokens
- Max output
- 29K tokens
- Lane
- Community: needs account opt-in; the provider may use data for training
- Modality
- Text
- Input price
- $0.6600 per 1M tokens
- Output price
- $1.00 per 1M tokens
- Processing region(s)
- us
- Route
- Via OpenRouter
- Cheapest plan
- Available from the Pro plan
- Price comparison
- Priced below 84 of 129 pro-band community models on Anoman (blended price: 1 part input to 3 parts output).
Released
2024-09
Training cutoff
2024-06
Parameters
32B (dense)
Ways to call
Ways to call Qwen 2.5 Coder 32B
Each model ID below is callable through the Anoman AI API. Price and processing region depend on the route.
| Model ID | Route | Input / output per 1M tokens | Region(s) | Lane | Context |
|---|---|---|---|---|---|
| openrouter/qwen-2.5-coder-32b | Via OpenRouter | $0.6600 / $1.00 | us | Community (opt-in, may train) | 33K |
| qwen-2.5-coder-32b-openrouter-backup | Via OpenRouter | $0.6600 / $1.00 | us | Community (opt-in, may train) | 33K |
Best for
Use cases
- IDE autocomplete + line completion
- Code review + bug detection
- Generating tests, docstrings, type annotations
- Translating code between languages
- Refactoring large diffs
Strengths
What it does well
- HumanEval 92.7% — matches GPT-4o-class coding ability
- Trained on 5.5T code tokens across 92 languages
- Native FIM (fill-in-the-middle) for IDE workflows
- Strong at long-form code understanding (LongBench-Code)
Limitations
Know the trade-offs
- Optimized for code — general chat is weaker than Qwen 2.5 72B
- Reasoning depth trails specialist models on math-heavy algorithm design
- 32K context cap vs Qwen 2.5's 131K
Benchmarks
Published scores
Scores from official model cards and public leaderboards. Higher is better unless noted.
| Benchmark | Score | Measures |
|---|---|---|
| HumanEval | 92.7 | Python code generation, pass@1 |
| MBPP | 82.0 | Mostly Basic Python Problems |
| LiveCodeBench | 31.4 | Recent contest problems (no benchmark contamination) |
| BigCodeBench | 38.3 | Realistic library-using tasks |
Code example
Chat completion with Qwen 2.5 Coder 32B
from openai import OpenAI
client = OpenAI(
base_url="https://api.anoman.io/v1",
api_key="anm-sk-..."
)
response = client.chat.completions.create(
model="openrouter/qwen-2.5-coder-32b",
messages=[{"role": "user", "content": "Hello!"}]
)
print(response.choices[0].message.content)Alternatives
Similar models in our catalog
Same maker first, then the closest price to Qwen 2.5 Coder 32B in the Anoman catalog.