LLM (Large Language Model)
A Large Language Model is a neural network trained on huge text corpora to predict the next token, capable of generating, summarizing, classifying, and reasoning over natural language.
LLMs are the engines under modern AI employees. Examples: GPT-5, Claude Opus 4.7, Gemini 2.5, Llama 4. The model takes a prompt (text in) and produces a completion (text out). Everything else — agents, tools, memory, employees — is software wrapped around this core capability. The tradeoffs you pick between models are: capability (hard reasoning vs. simple tasks), latency (chat-fast vs. batch), cost (cents per million tokens), and context window (how much can fit in a single prompt). A production stack usually uses several models routed by task.
Example
An AI Bookkeeper might use a fast cheap model for OCR-fixup on every invoice, a mid-tier model for categorization, and a top-tier model only when escalating an ambiguous case for a final classification.
How OpenLabor uses it
OpenLabor routes between models per task — using premium models only when capability matters and cheaper models for everything else, so you pay for results not for compute.
Which LLM should I use?
It depends on the task. For most production agentic work in 2026, the answer is a mix: a frontier model for reasoning steps, a cheaper one for high-volume simple steps.
Are open-source LLMs good enough?
For many tasks yes — Llama and Mistral families have caught up. For complex multi-step agents, frontier closed models still lead by a real margin.
Related: rag, tool-use, fine-tune, token-cost.
AI Labor Glossary