Cerebras Inference is a cloud service from Cerebras Systems, a company that designs its own AI chip, the Wafer-Scale Engine. Inference means running an already trained model to produce an answer, as opposed to training it. The chip is built from one whole silicon wafer, and Cerebras says it is 58 times larger than a GPU and runs inference up to 30 times faster than GPU systems.
A developer sends chat messages to the API and receives the model's reply; Cerebras publishes a speed of about 3,000 tokens per second for GPT OSS 120B. The API is compatible with OpenAI's API, so existing code needs only small changes, and Python and Node.js SDKs exist. The documentation lists streaming, structured outputs, tool calling (the model asks the program to run a function) and prompt caching. Companies that need fixed capacity can rent dedicated inference with reserved hardware.
Cerebras Inference is a hosted service. A free trial gives a small credit; the Developer tier is pay per token (a token is a piece of a word); Enterprise is by agreement. AWS Marketplace, OpenRouter and Hugging Face also offer it.