Recommendation #1 (The Powerhouse)
ollama pull qwen2.5-coder:3b

Recommendation #2 (The 2026 New Arrival)
ollama pull qwen3-coder:4b - not working

ollama pull qwen3.5:4b

^^^ working

for coding use this one as its trained more on code

ollama pull qwen2.5-coder

To use it in Open WebUI:

1. Once the download finishes in your terminal, go to your Open WebUI browser tab.

2. Refresh the page.

3. Click the model selector at the top (where it says Llama 3.2).

4. Select Qwen2.5-Coder:3b from the list.

⚠️ Pro-Tip for 4GB Coding

When using a coding model, the "Context Window" (how much code the AI can "read" at once) is what eats up your remaining VRAM.

If you paste a 500-line script and the model crashes, it's because the "KV Cache" filled up your last 1GB of VRAM. To fix this in Open WebUI:

1. Go to Settings -> Advanced Parameters.

2. Find Context Length.

3. Set it to 4096 instead of the default 8192 or 32k. This limits how much "history" the AI remembers but guarantees it won't crash your 1650 Super during a long coding session.