BTC
ETH
SOL
XRP
BNB
TON
DOT
LTC
POL
HBAR
NEAR
ARB
UNI
The moment an AI model is used to answer: it takes your input and works out the output. Each answer you get is one inference.
AI Drafted with AI help · Reviewed by Keshav Aggarwal · Updated
In simple words
Learning to cook takes months. Cooking dinner tonight takes 20 minutes.
For AI, the learning part is training. Making tonight’s dinner, answering your question, is inference.
The technical side
Inference is running a trained model on new input to produce an output. For language models it happens token by token, mostly on GPUs or other AI chips, and its cost grows with the number of input and output tokens. That is why API providers price per million tokens.
Training a big model is a one-off cost; inference is paid every time the model is used, so cheaper and faster inference is a major focus of AI companies. Compare inference prices on our LLM API prices page.
Inference kya hai?
Inference matlab AI ko use karna: tumne sawaal poocha, model ne jawab nikala. Training ek baar hoti hai, inference har sawaal pe hota hai.
Related terms
- LLM (Large Language Model)AIAn AI model trained on huge amounts of text so it can read and write language, like the models behind ChatGPT, Claude and Gemini.
- TokenAIA small piece of text, often part of a word, that an AI model reads and writes. AI APIs charge by the number of tokens.
- GPU (Graphics Processing Unit)AIA computer chip that does thousands of small calculations at once. First made for video games, it now trains and runs most AI models.