Choose a model for the task
- Chat, drafts, and short questions. Start with a small variant such as Qwen3.5 2B.
- Photos and screenshots. Filter for Vision in the model catalog and check that your variant supports images.
- Agent work and web search. Look for a model that can call tools. Use the Tool use filter in the catalog.
Compare download size and context
The download size shows how much space the files take up. The model needs additional memory while it runs. A longer context can require more memory.
Models can come in several versions with different weight
precision. Labels such as Q4_K_M and
Q8_0 identify these compressed variants. For the same
model, Q4 usually takes up less space than Q8. Compare the actual
file sizes and try a smaller variant if speed matters.
If a model starts but takes a long time to reply, reduce the context size in Settings → Models and API. The model page lists a variant’s context limit when it is known.
Install with Ollama or Hugging Face
- Ollama. Copy the Ollama variant name from the model page. In codef, open Models and downloads → Installed → Install by Ollama name and paste it.
- Hugging Face. Open the Hugging Face tab, search for the repository or paste its link, choose a GGUF file, and select Download.
- A file on your computer. From Installed, select Import GGUF and choose the file.
On Windows and Linux, install and start Ollama first. Some models require a newer version. The model page shows the required Ollama version when known.
Track downloads and start chatting
The Downloads tab lists transfers separately. It shows progress, speed during active transfers, and the import stage. You can cancel a transfer or retry a failed one.
Once the model appears under Installed, select Use and send a short message. If the download stalls, check disk space and repository access.