Guides · Generative AI

Ollama Coding Models: Run Qwen2.5-Coder Locally with a Coding Agent

By Ravindra Bagale, trainer · LinkedIn · Published: 25 October 2026 · Last updated: 25 October 2026 · 4 min read

Ollama runs AI models on your own computer, so code never leaves your machine. As of 25 October 2026, the Ollama library lists qwen2.5-coder in 0.5b, 1.5b, 3b, 7b (about 4.7 GB, the default), 14b (about 9 GB) and 32b (about 20 GB), and deepseek-coder-v2:16b. Run ollama run qwen2.5-coder to chat. To use it inside a coding agent, Ollama documents ollama launch opencode --model qwen2.5-coder (and similar for Claude Code). Small local models are weaker than cloud models, so use them for simple tasks.

1. Why: privacy, offline, no per-token bill

Local models keep code on your machine, work without internet after download, and do not need a paid plan. The trade-off is speed and quality on small hardware.

2. What: model sizes

3. How: set it up

1. Install Ollama from ollama.com (Windows, macOS or Linux).
2. Choose a size: start with 7b on 16 GB RAM; 1.5b or 3b on 8 GB.
3. Run ollama run qwen2.5-coder and ask a question.
4. Call it from Python using the local API at localhost:11434.
5. Optionally run ollama launch opencode --model qwen2.5-coder.

4. Source

• ollama.com/library/qwen2.5-coder tags and sizes (checked 25 Oct 2026).
• Ollama docs for ollama launch with coding tools.
• Sizes are downloads; RAM needed is higher.

5. Worked example 1: pull and chat (start here)

In a terminal:

ollama run qwen2.5-coder "Write a Python function is_palindrome(s) with 2 tests."

Expected output (yours will be similar, not the same):

def is_palindrome(s: str) -> bool:
    s = "".join(c.lower() for c in s if c.isalnum())
    return s == s[::-1]
...

The first run downloads about 4.7 GB.

6. Worked example 2: call it from Python

pip install requests, then save as ask.py:

import requests

r = requests.post("http://localhost:11434/api/generate", json={
    "model": "qwen2.5-coder",
    "prompt": "Explain what a Python list comprehension is in 2 lines.",
    "stream": False,
})
print(r.json()["response"])

Expected output (yours will be similar, not the same):

A list comprehension builds a new list in one line,
for example [x*2 for x in nums].

7. Worked example 3: connect a coding agent

With OpenCode installed:

ollama launch opencode --model qwen2.5-coder

Expected output (yours will be similar, not the same):

OpenCode opens in your terminal using the local qwen2.5-coder model.

Ollama docs also show launching Claude Code against a local model; check the docs for setup notes.

Names, amounts and companies in the examples are made up for practice.

8. Common mistakes

9. Try at home

Recap

Local is private; pick a size your machine can carry.

Related: Run an LLM Locally with Ollama on Windows and Mac (Beginners)