LLM
Large Language Model — a language processing model trained on massive amounts of text, capable of generating coherent text and reasoning about instructions.
Definition
An LLM (Large Language Model) is a neural network trained on massive text corpora to predict and generate text token by token. Modern LLMs (GPT-4, Claude Sonnet, Gemini, DeepSeek) have learned to understand and generate code, follow complex instructions, and reason through multi-step problems. For developers, an LLM is the reasoning engine of a coding agent: it decides what to do, the coding agent is the environment that gives it the tools to do it.
postcursors perspective
The distinction between "the LLM" and "the coding agent" is important to maintain. Claude Sonnet 4.5 is an LLM. Claude Code is a coding agent that uses Claude Sonnet 4.5 as its engine. OpenCode is a multi-provider agent that can use Sonnet, DeepSeek V3, Gemini Flash, or other LLMs. This distinction explains why switching LLMs within the same agent changes results without changing workflows.
In practice
In an agentic workflow: the LLM receives the context (instructions, code, test results), decides on actions to take via tool use, and generates the modifications. Switching LLMs (e.g., DeepSeek V3 → Claude Sonnet) within the same agent changes the quality of reasoning and the cost, not the workflow or configuration.
Common misconceptions
- ✗ Using 'ChatGPT' to refer to LLMs in general — ChatGPT is an OpenAI product, GPT-4o is the underlying LLM
- ✗ Confusing the LLM with the agent — Claude Code ≠ Claude Sonnet