12 Comparing and selecting LLM models
This chapter covers
- The five major open-weight model families: Llama, Gemma, Qwen, Mistral, and DeepSeek
- How model size affects quality, speed, and memory requirements
- Which models work best for different languages
- Switching models in your Streamlit chatbot
- The February 2026 revolution: frontier open models that match proprietary systems
When you first visit the Ollama model library, the sheer number of available models can feel overwhelming: there are dozens of names, cryptic version numbers, and parameter counts ranging from half a billion to hundreds of billions. How do you choose?
This chapter will give you a practical framework for understanding and selecting LLM models. By the end, you will know the major model families, understand the tradeoffs between size and quality, and be able to switch between models in your chatbot with ease.
Note Model names and rankings change quickly. Treat the tables in this chapter as a snapshot, and learn the model-checking workflow: run ollama list to see what you have, run ollama pull <model> to download a model, and check the Ollama model library for the latest tags before choosing one.