GLOSSARY
Inference
The act of running a trained model to produce an output — every chat reply, generated image or API call is one inference pass.
Inference is the production phase of AI: using a trained model to generate outputs, as opposed to training, which builds the model. Every message you send to Claude, every image Midjourney renders, is inference — and it costs real compute, which is why usage limits and per-token pricing exist.
Inference efficiency is a quiet arms race: faster chips, quantized models and clever serving let labs deliver frontier quality at falling cost. The 'reasoning models' trend extends inference inside the model — spending more compute thinking before answering to improve hard problems.