DeepInfra is an AI inference cloud, a hosted service that runs machine learning models and returns their output to programs over an API. Inference is the step in which a trained model turns an input into an output. The company was founded in September 2022 and runs hundreds of open-source models on its own hardware in US data centers.
A developer sends requests to an OpenAI-compatible API, so code written for OpenAI's SDK works after a change of the base URL. The catalog covers large language models such as DeepSeek, Qwen, Kimi and GLM, as well as embeddings and rerankers, image and video generation, speech recognition and text-to-speech. Under its zero-retention policy, inputs and outputs are not kept; the company is SOC 2 and ISO 27001 certified. Customers can also deploy their own fine-tuned models on dedicated GPUs or rent GPU clusters.
Pricing is pay as used: language models per million input and output tokens, most other models by inference execution time, with no long-term contract.