Baseten is an inference platform. Inference means running a trained AI model to produce answers, as opposed to training it. Baseten runs open-source, fine-tuned and custom models on its own GPU infrastructure, which it calls the Baseten Inference Stack, and serves them to other software over the internet.
It has two main offers. Model APIs give instant access to pre-optimized open models such as GLM, DeepSeek, Kimi and gpt-oss: a program sends text and receives the model's output, priced per token. Dedicated deployments run a customer's own model on GPUs reserved for it, with autoscaling (adding and removing servers as traffic changes) and fast cold starts. Baseten also serves image generation, transcription, text-to-speech and embedding models, and offers training jobs. Models run in Baseten Cloud, in a customer's own VPC (virtual private cloud), or both.
It is a hosted service billed by use: per million tokens for Model APIs and per minute of GPU time for dedicated deployments, with no charge for idle time.