We may earn a commission if you buy through links on this page. It never changes a score or a ranking. How we make money
There's a free plan; paid tiers are quoted.
Baseten's Pay as you go (Basic) plan has a $0 base platform fee, and the vendor offers starter credits with an unpublished amount and duration. Usage is billed on demand, so cost depends on the models and compute resources you run. Sandbox CPU, RAM and snapshots have listed rates, while dedicated inference and training rates vary by hardware. Model APIs are billed per 1 million tokens. Production scale, Pro and Enterprise require custom quotes.
01Every plan, line by line
| Plan | Billed annually | Billed monthly | Built for |
|---|---|---|---|
| Pay as you go (Basic) | — | $0 | Developers starting with model inference |
| Production scale | — | Quote | Teams needing reserved production capacity |
| Pro | — | Quote | Teams scaling dedicated model serving |
| Enterprise | — | Quote | Organizations with security and hosting needs |
Add-ons and overages
- Sandbox usage: CPU $0.0655/hr, RAM $0.0211/GB-hr, snapshots $0.20/GB-mo; inference and training rates vary by hardware from $0.01052 to $0.16633/min; CPU starts at $0.00058/min; Model APIs are billed per 1M tokens. Overages and add-on prices are not published.
02What the plan grid doesn't tell you
Token rates
Model APIs are billed per 1 million tokens, but token rates are not listed here. Get prices for each model and expected input/output mix.
Hardware choice
Dedicated inference and training range from $0.01052 to $0.16633 per minute, depending on hardware. Request rates for your target configuration.
Sandbox and snapshots
Sandbox CPU costs $0.0655 per hour, RAM $0.0211 per GB-hour and snapshots $0.20 per GB-month; include them in workload estimates.
Unpublished overages
Overages and add-on prices are not published. Ask how usage limits, extra capacity and any support or reserved compute affect your bill.
03Which plan should you buy?
| If you are… | Get | Why |
|---|---|---|
| Prototype or side project | Pay as you go (Basic) | The $0 base platform fee suits starting on demand, but budget for compute, tokens and snapshots. |
| Production team needing priority | Pro | Pro lists dedicated compute, unlimited autoscaling, priority access to high-demand GPUs and higher Model API rate limits; pricing is custom. |
| Organization with VPC requirements | Enterprise | Enterprise includes self-hosted or hybrid deployment options, custom SLAs, global regions and data residency controls; request a quote. |
04Is Baseten worth the money?
Basic is a practical way to start because it has no base platform fee, but it is not a fixed-cost free tier: compute, API tokens and snapshots add usage charges. It can suit prototypes or workloads that can scale with demand. For production, get a quote that models your hardware, token volume, traffic peaks and support needs. Self-hosting is an Enterprise option, not a free route around platform costs, because Baseten still manages the control plane and your organization operates the workload infrastructure.
Baseten pricing FAQ
Does Baseten have a free plan?
Pay as you go (Basic) has a $0 base platform fee, with usage billed on demand. Baseten also offers starter credits, but their amount and duration are not published. Compute and other usage can still incur charges, so confirm the applicable limits and estimate your workload before deploying.
How much does Baseten inference cost?
Dedicated inference rates range from $0.01052 to $0.16633 per minute, depending on hardware. Model APIs are billed per 1 million tokens, but token prices are not listed here. Ask Baseten for current rates by model and hardware, using your expected request and token volumes.
How much do Baseten Sandboxes cost?
The listed Sandbox rates are $0.0655 per CPU hour and $0.0211 per GB-hour of RAM. Snapshots cost $0.20 per GB-month. Estimate CPU time, memory use and stored snapshot volume separately when calculating a Sandbox workload's cost.
Are Baseten's production plans fixed-price?
Production scale, Pro and Enterprise require custom quotes, and volume discounts are available. No annual prices are published. Ask for a quote that covers reserved capacity, model and compute usage, regions, support, rate limits and any overage terms.