AI basics
Why it matters
Instead of answering straight away, a reasoning model spends extra computing time working a problem through, much like showing your work in math class. It tends to do better on puzzles, math and code, and worse on speed and cost.
Model makers often offer both a fast mode and a reasoning mode. When a release says a model thinks longer, it usually means more inference compute per answer, which is one reason the demand for AI chips keeps growing after training is done.
Related terms
- Inference
- Inference is running an already-trained AI model to produce an answer, such as a chatbot reply. It is separate from training, which builds the model, and it runs for every user request.
- Large language model (LLM)
- An AI model trained on huge amounts of text to predict the next word, which lets it answer questions, write and code. Grok is a family of LLMs.
- AI agent
- An AI that takes actions toward a goal, such as clicking, filling in forms or running code, instead of only replying. Grok Bot is an agent.