Model
The trained network itself: billions of parameters that turn a context into the next token, and nothing more.
A model is the trained neural network: a very large set of parameters that, given a sequence of tokens, produces a probability for what the next token should be. That's the entire job. It doesn't read your repo, doesn't remember yesterday, doesn't run tests. Every one of those abilities is added by the harness around it.
This is why the same model behaves so differently in different products. Claude Sonnet in a chat window answers questions; the same model inside Claude Code edits files and runs your test suite. The model didn't change; what changed is the system prompt, the tools on offer and the loop that feeds results back in. When behaviour shifts between two products, suspect the wrapper before the weights.
The model is also stateless: it holds nothing between calls. Each request starts from a blank slate and sees only what the harness sends. When you pick a model in a dropdown (Sonnet vs Opus, GPT-5 vs a mini variant) you're trading capability against speed and price, but you are never changing what it can perceive. That's the harness's job.
- Claude Code
/modelswitches between Claude models mid-session; everything else about the session stays the same. - CodexThe model is set in config or the model picker; the CLI is the harness that gives it a sandbox and tools.
- OllamaRuns open-weight models locally; the same model file behaves differently depending on which harness calls it.
“Should I switch models? It keeps missing the config file.”
“It can't see the config file. Same model, different context, and it'd be fine. Point it at the path first.”
Course 01 puts every one of these terms to work: you install Claude Code, run the loop, and ship a real project — permission modes, compaction, hooks and all.
Start Course 01 →