Baseten AI features and pricing (October 2026)

Baseten is an inference platform for teams that run open-source, fine-tuned or custom AI models in production. It provides hosted Model APIs, dedicated GPU deployments with autoscaling and training jobs, either in Baseten's cloud or in the customer's own environment.

The platform covers three paths. Model APIs expose a curated set of hosted models through an OpenAI-compatible API, with support for the Anthropic Messages API in beta. Dedicated deployments run an open-source, private or custom model on reserved GPUs, configured through YAML files, Python classes or custom Docker containers. Training offers Loops for fine-tuning and training jobs that run custom code on GPU clusters.

Baseten manages model containers, GPU capacity across clouds and regions, scaling and observability. Documented hosting options are Baseten Cloud, single-tenant and self-hosted setups, and the Enterprise plan adds hybrid use with the customer's VPC. Baseten Labs, Inc. is based in San Francisco. Typical users are engineering and platform teams that build AI features into their products.

• Serves models with autoscalingDedicated deployments add or remove replicas based on concurrent requests per replica, within a minimum and maximum you set. With the default minimum of zero, all replicas shut down when no traffic arrives, so platform teams pay for GPU minutes only while a model is in use.
• Compiles models for faster inferenceEngine-Builder-LLM compiles dense text models with TensorRT-LLM, BIS-LLM serves large mixture-of-experts models with distributed inference, and BEI handles embedding, reranking and classification models. Teams can let Baseten choose the engine or run vLLM or SGLang in a custom container.
• Calls hosted models through one APIModel APIs expose hosted open models such as GPT OSS 120B through an OpenAI-compatible endpoint and bill per million tokens. Developers call them with an API key and the OpenAI SDK, which suits prototypes before a team commits to a dedicated deployment.
• Fine-tunes and trains modelsLoops provides dedicated trainers and samplers for supervised fine-tuning and reinforcement learning on supported models, and training jobs run custom code on GPU clusters. The resulting checkpoints can be served on Baseten, so teams avoid moving weights to a separate serving platform.

• GPU time is billed per minute, from $0.01052 for a T4 to $0.16633 for a B200, as of September 2026.
• The Enterprise plan supports self-hosting in the customer's VPC, hybrid setups, custom global regions and control over data residency.
• Model APIs accept the OpenAI Chat Completions format and, in beta, the Anthropic Messages API.
• The pricing page lists SOC 2 Type II certification and HIPAA compliance for every plan, including Basic.

Baseten Labs, Inc. is based in San Francisco, California, according to its privacy policy. Enterprise customers can arrange custom regions and self-hosting.

Check which GPU types and regions your workloads need, because the listed prices apply to standard capacity and other countries and regions go through sales. For EU personal data, review the public DPA, the subprocessor list in the trust center and the transfer basis, which relies on Standard Contractual Clauses. Compare per-minute dedicated pricing with per-token Model APIs for your expected traffic, and note that custom SLAs are part of Enterprise only.

No reviews yet. Be the first to share how your team uses it.

Every review shows its verification level. We read each one before it goes live. How we check reviews