Baseten

What it does

Baseten takes in open-source, custom, or fine-tuned AI models, or requests against hosted models, and runs them for inference on managed GPU infrastructure. Teams can call supported hosted models through an OpenAI-compatible API, deploy their own models on dedicated GPUs with the Baseten CLI, or train models with Loops or Training Jobs and then deploy the resulting checkpoints. The output is live model endpoints with logs, metrics, request traces, and autoscaling.

People typically start with Model APIs to test a supported model with a single API call, then move to dedicated deployments when they need custom models, specific hardware, or their own autoscaling settings. Baseten packages models with its open-source Truss standard, and its inference engines (such as Engine-Builder-LLM and BEI) optimize supported architectures. Capacity can be provisioned across clouds and regions, and Baseten can also run inside a customer's own VPC.

It suits engineering teams building AI products that need production-grade serving, and model labs distributing their models. Accounts start with credits, and Model APIs use token-based pricing while dedicated compute bills by the minute. Pro and Enterprise plans are quote-based, and some features require talking to sales. Limitations include that Model APIs cover only a curated set of models, and regional compute outside listed regions requires contacting sales.

What it can do

• Serve hosted language, embedding, and speech models through OpenAI-compatible and Anthropic Messages APIs (Anthropic in beta)
• Deploy open-source or custom models on dedicated GPUs with a config file, Python model class, or custom Docker server
• Autoscale deployments with fast cold starts and per-minute compute billing
• Fine-tune models with Loops or run custom training code with Training Jobs, then deploy checkpoints
• Export logs, metrics, and request traces to Datadog, Prometheus, Grafana, or New Relic
• Self-host Baseten in your own VPC or use hybrid capacity

Works with

• Datadog
• Prometheus
• Grafana
• New Relic

Example workflows

Call a hosted model from an app

Illustrative example. Input: a supported model name and a chat prompt. Steps: create an API key, point the OpenAI SDK at Baseten's Model APIs endpoint, send the request. Output: a generated completion returned through the standard Chat Completions response.

Deploy a custom model on dedicated GPUs

Illustrative example. Input: a model's weights and a config.yaml specifying hardware and engine. Steps: install the Baseten CLI, deploy with the CLI, test the endpoint, promote the deployment to production. Output: a dedicated inference endpoint with autoscaling and request logs.

Before you choose

• Model APIs offer selected models; custom models require a dedicated deployment.
• Compute can accrue during model loading and scaling; verify autoscaling and scale-to-zero settings.
• Hardware, regions, VPC deployment and support depend on the selected agreement.

Listed under

• Coding
• Sandboxes and infrastructure
• Developers
• GPU infrastructure
• Inference
• LLM serving
• ML engineers
• Model APIs

Get started

New accounts receive credits. Model APIs can be called with an API key and the OpenAI SDK.

1. Create an account at login.baseten.co/sign-up and note the credits for initial experiments.
2. Generate an API key in the workspace and call a Model API using the OpenAI SDK, following the quickstart.
3. For dedicated deployments, install the Baseten CLI, define the model in config.yaml, and deploy.

Basic has no platform subscription fee; model and compute usage are billed separately. Model APIs meter model-specific input, cached-input and output tokens. Dedicated compute is billed by the minute, including model loading and scaling activity. New accounts receive introductory credits; Pro and Enterprise terms are quoted.

Source: baseten.co/pricing, checked 29 Oct 2026. Prices can change; check before you buy.

Sources for this overview

Product pages, documentation, and pricing checked 29 Oct 2026. Each source supports part of the overview; availability and prices can change.

• Inference Platform: Deploy AI models in production | Baseten ↗www.baseten.co
• Cloud Pricing ↗www.baseten.co
• Baseten overview - Baseten ↗docs.baseten.co

Company website

baseten.co: product, pricing and policy pages

YC directory mirror

Batch and company name

Social accounts

Accounts the company links, and its posts on X

Human review

Checked by a person

knowledge_base

Last checked 29 Oct 2026. Every field keeps its source and date; unknown means we couldn’t confirm it.