Baseten
Inference platform for open and custom models
Last reviewed Sep 28, 2026
Baseten is an inference platform that serves popular open models through per-token Model APIs and runs custom or fine-tuned models on dedicated GPU deployments billed per minute, with managed training jobs on the same hardware.
Available GPUs
Hourly on-demand pricing. Click column headers to sort.
Prices last updated: October 6, 2026
Baseten pricing by GPU
Configurations, price rank, and alternatives for one GPU at a time.
LLM API Pricing
Pay-per-token pricing. Prices shown per 1M tokens.
Prices last updated: October 6, 2026
| Model | Input/1M | Output/1M |
|---|---|---|
| $0.100 | $0.500 | |
| $0.130 | $0.260 | |
| $0.150 | $0.500 | |
| $0.300 | $1.20 | |
| $1.32 | $3.96 | |
| $1.40 | $4.40 | |
| $1.40 | $4.40 | |
| $3.00 | $15.00 |
Baseten pricing by model
Input, output, and batch rates, plus alternatives, for one model at a time.
Pros & Cons
Advantages
- Per-minute billing on dedicated GPUs
- Per-token Model APIs and dedicated deployments from one account
- SOC 2 Type II certified and HIPAA compliant
- Multi-GPU instance sizes published with vCPU and memory
Limitations
- Dedicated GPU rates are higher than raw GPU rental because they include the serving platform
- Model API catalog is a curated set of models
- Volume discounts and custom regions require a sales conversation
Key Features
Model APIs
OpenAI- and Anthropic-compatible endpoints for hosted open models, billed per input, cached input, and output token
Dedicated deployments
Custom, fine-tuned, or open-source models on dedicated GPUs from T4 to B200 and RTX PRO 6000, in 1 to 8-GPU instances, with autoscaling
Training
Managed training jobs on the same GPU instance types, with checkpoints deployable to inference
Truss
Open-source packaging standard for serving models built in any framework
Self-hosted and hybrid
Enterprise deployments in the customer's VPC or split across Baseten's cloud and the customer's
Pricing Options
| Option | Details |
|---|---|
| Model APIs | Per 1M input, cached input, and output tokens |
| Dedicated deployments | Per minute of GPU instance time, scaled to zero when idle |
| Pro and Enterprise | Volume discounts, priority GPU access, and self-hosted options |
Availability & Support
Support
Email and in-app chat on Basic; dedicated Slack and Zoom support on Pro and Enterprise
Company
Published by Baseten to identify the company behind this listing.
- Headquarters
- 🇺🇸 San Francisco, United States
Getting Started
- 1
Create an account
Sign up; new accounts include credits
- 2
Call a Model API
Use an OpenAI-compatible client against a hosted model
- 3
Deploy a model
Package a model with Truss or pick one from the model library and choose a GPU instance type
Frequently Asked Questions
What GPU types does Baseten offer?
Baseten offers various GPU types including Tesla T4, Tesla T4, Tesla T4, A100 SXM, A100 SXM, A100 SXM, A100 SXM, A100 SXM, H100 SXM, H100 SXM, H100 SXM, H100 SXM, H200, H200, H200, H200, A10, A10, A10, A10, B200, B200, B200, B200, L4, L4, L4, RTX PRO 6000, RTX PRO 6000, RTX PRO 6000, RTX PRO 6000. Check the pricing table above for current availability and pricing.
How do I get started with Baseten?
Create an account, Call a Model API, Deploy a model
What are Baseten's main advantages?
Baseten's main advantages include: Per-minute billing on dedicated GPUs, Per-token Model APIs and dedicated deployments from one account, SOC 2 Type II certified and HIPAA compliant, Multi-GPU instance sizes published with vCPU and memory.
What are Baseten's limitations?
Baseten's main limitations include: Dedicated GPU rates are higher than raw GPU rental because they include the serving platform, Model API catalog is a curated set of models, Volume discounts and custom regions require a sales conversation.
Compare Providers
Find the best prices for the same GPUs and models from other providers
