Openhearth

2026-11-15

Which open-source AI model should you use?

Start from the job, not from a leaderboard. A smaller Llama model is enough for short notes and quick rewrites. A larger Llama or Qwen model is a better fit when the answer needs to stay coherent for several paragraphs or when you are editing code. Mistral Small is a calm default for summaries. Gemma is useful when you want a concise reply.

Then read the licence. Llama models ask you to show “Built with Llama” and to follow Meta’s acceptable use policy. Gemma has its own terms. Mistral Small and the Qwen weights listed here are commonly under the Apache Licence 2.0, which you should confirm on the model card before a public launch.

If the chat must stay on a machine you control, run the same families through Ollama and point this app at that server. The model is still the model. Only the computer changes.