Skip to main content

Glossary Term

LLM (Large Language Model)

By The Codegen Team · Updated March 26, 2026

A neural network trained on massive text datasets to predict and generate human-like text, powering all modern AI coding tools.

A large language model is a neural network trained on massive text datasets to predict and generate human-like text. LLMs power all modern AI coding tools, from inline autocomplete to autonomous agents.

The major LLM families used in coding tools include GPT (OpenAI, used by Copilot), Claude (Anthropic, used by Claude Code), Gemini (Google), and various open-source models like Llama and Mistral.

Model size, training data, and fine-tuning determine what a model can do. Larger models with code-specific training generally perform better on complex programming tasks, but smaller models can be faster and cheaper for routine completions.

In plain English

The AI system underneath tools like Claude, ChatGPT, and Cursor — trained on large amounts of text, it learned to understand and generate language, including code.

Why it matters

Every AI coding tool is powered by an LLM underneath. Understanding LLMs helps you interpret benchmark results, understand why context window size matters, recognize why tools occasionally invent APIs that do not exist, and evaluate which tools are likely to perform better on reasoning-heavy tasks versus pattern-matching tasks. The model is the engine — the tool built on top of it is the vehicle.

In practice

Two developers use different AI coding tools on the same refactor task. One tool uses GPT-4o; the other uses Claude. Both produce working code, but the Claude-powered tool produces a more architecturally coherent result with fewer edge cases missed. The difference is not the interface — it is the underlying model's reasoning capability on multi-file, multi-step tasks. SWE-bench exists precisely to measure this difference objectively.

How Codegen uses LLM (Large Language Model)

Codegen runs on Claude and does not expose the LLM directly. The model provides the reasoning; Codegen provides the structure around it — task context from ClickUp, sandboxed execution, audit trails, cost tracking, and the PR review layer. Teams choosing Codegen are choosing the orchestration infrastructure, not a specific model. For enterprise deployments that require a different model — a privately hosted one, or a fine-tuned variant — Codegen supports custom model configurations. The LLM is a component, not the product.

Frequently Asked Questions