Model

What is a Model

The model is the underlying trained system, Claude itself, that receives a block of context and produces the next piece of output: text, or a structured request to call a tool. On its own, a model cannot read a file, run a command, or reach the internet, it can only generate what it predicts should come next given what it was shown.

Everything that turns that raw generation into useful action comes from the harness wrapping it: assembling the context, executing whatever tool calls the model requests, feeding results back in. This is why the exact same underlying model can behave very differently in two different tools, the model's own reasoning is one input, but the harness decides what it actually gets to do with that reasoning.

The model is not the same thing as the agent built around it, an agent is the model plus the loop, tools, and harness that let it act repeatedly toward a goal. A model alone, without any of that scaffolding, is closer to a single, very capable autocomplete than to something that can independently get a task done.