Docs · Guide I
Which model for which job
Every job starts on the local daily driver. A different model only takes over when the job has a clear reason for it.
The rule: start local
qwen3.8:27b is the default for everyday chat, drafting and questions. The other four models each have one specific job. If a task doesn't match one of those jobs, it stays on the daily driver.
Job to model
| Job | Model | Runs |
|---|---|---|
| Everyday chat, drafting, questions | qwen3.8:27b | Local |
| Voice control in Home Assistant | qwen3.5:9b | Local |
| Searching notes by meaning | embeddinggemma-2 | Local |
| Hard coding problems | glm-5.3:cloud | Ollama Cloud |
| High-volume batch work, where cost per run matters | glm-5.3-flash:cloud | Ollama Cloud |
Decide in five questions
- Is it a voice command for Home Assistant? Use
qwen3.5:9b. It's smaller, so replies come back quickly. - Is it finding something in the notes? Use
embeddinggemma-2. It's an embedding model: it turns text into vectors for search. It doesn't chat. - Is it a hard coding problem? Use
glm-5.3:cloud. - Is it the same small job, many times over? Use
glm-5.3-flash:cloud, the low-cost option for bulk work. - Anything else? Use
qwen3.8:27b.