LLM (Large Language Model)

A large language model is a neural network, typically a transformer architecture, trained on hundreds of billions to trillions of tokens of internet text. The training objective is simple: given previous tokens, predict the next one. Scaled up sufficiently, this single capability produces models that can summarize, translate, answer questions, write code, and reason.

LLMs as of 2026 include: OpenAI's GPT-4 / GPT-4.5 family, Anthropic's Claude 4 family (Opus, Sonnet, Haiku), Google's Gemini 2.5 Pro / Flash, Meta's Llama 4, and many specialized open models. Frontier model performance has converged in 2025-2026 — choice often comes down to context window, latency, cost, and specific task performance.

For Salesforce: an LLM is the "AI engine" inside Agentforce, Einstein Copilot, and platforms like gptfy. The LLM doesn't store customer data — it processes prompts and returns responses. Architecture decisions about which LLM, how to send prompts, and how to handle responses define the entire AI experience.