Baseten

Baseten is an inference platform for deploying and serving AI models in production. It runs open source, custom and fine tuned models on infrastructure built for high performance inference, using its own Baseten Inference Stack with custom kernels, modern decoding techniques and advanced caching, and it offers cross cloud high availability with fast cold starts. It is aimed at AI product companies and engineering teams that need fast, reliable model serving at scale. Dedicated deployments run a team's own models on dedicated GPUs, Model APIs give instant access to pre optimised popular models, and Training lets teams train models, including with reinforcement learning through the Loops SDK, then deploy them on the same stack. Baseten for Model Labs helps model developers distribute and monetise their models. Workloads can run in Baseten Cloud, including single tenant clusters, self hosted in the customer's own VPC, or in a hybrid of both. The platform is SOC 2 Type II and HIPAA compliant, and forward deployed engineers are available to help optimise models. Basic costs $0 per month, pay as you go, and includes dedicated deployments, Model APIs, training and email and chat support. Pro adds priority GPU access, dedicated compute, higher rate limits and Slack and Zoom support; Enterprise adds custom SLAs, self hosting, data residency control and advanced RBAC. Both are quote based with volume discounts. Model API prices are published per million tokens, for example Kimi K3 at $3.00 input and $15.00 output, and GPT OSS 120B at $0.10 input and $0.50 output. Dedicated GPUs are billed by the minute, such as T4 at $0.01052, L4 at $0.01414 and A10G at $0.02012 per minute. Baseten has announced that Blaxel is joining the company.