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, but it is not free to run workloads: inference, training and sandbox use are billed separately. The pricing page publishes some compute rates, including sandbox and selected GPU rates, but not Model API token prices. Production scale, Pro and Enterprise are custom-quoted, and annual pricing is not published. We’d treat the $0 entry as a low-friction starting point, not a predictable monthly bill.
01Every plan, line by line
| Plan | Billed annually | Billed monthly | Built for |
|---|---|---|---|
| Pay as you go (Basic) | — | $0 | Teams starting with usage-based inference |
| Production scale | — | Quote | Teams needing reserved production capacity |
| Pro | — | Quote | Teams scaling model deployments |
| Enterprise | — | Quote | Organizations with security and deployment needs |
Add-ons and overages
- Dedicated Inference and Training rates range from $0.01052/min on one thread to $0.16633/min on B200; other listed rates: L4 $0.01414/min, A10G $0.02012/min, H100 MIG $0.0625/min, A100 $0.06667/min, H100 $0.10833/min.
- Sandbox snapshots: $0.20/GB-mo.
- Overage costs and seat costs are not published; usage is billed on demand without overage penalties.
02What the plan grid doesn't tell you
Unpublished Model API rates
Model APIs are billed per 1M tokens, but rates are not listed. Get model-specific pricing before estimating spend.
GPU rates vary
Dedicated Inference and Training range from $0.01052/min on one thread to $0.16633/min on B200; other GPU rates vary.
Sandbox usage and snapshots
Sandbox CPU costs $0.0655/hour and RAM $0.0211/GB-hour; snapshots add $0.20/GB-month.
Unpublished seat and annual prices
Annual pricing and seat costs are not published. Usage is billed on demand without overage penalties; custom plans may offer volume discounts.
03Which plan should you buy?
| If you are… | Get | Why |
|---|---|---|
| Teams starting with inference | Pay as you go (Basic) | It has a $0 base platform fee and usage-based billing across Model APIs, Dedicated Inference, Training and Sandboxes. |
| Teams reserving production capacity | Production scale | It offers reserved and priority compute, custom capacity and regions, and dedicated support at a custom quote. |
| Organizations needing deployment controls | Enterprise | It includes custom SLAs, advanced security and compliance, self-hosted or hybrid deployment, and data residency controls. |
04Is Baseten worth the money?
Baseten may suit teams that need managed model serving and want to start without a platform fee. The catch is budget predictability: Model API rates are not listed, while paid plans require custom quotes. Published GPU and sandbox rates help with some estimates, and custom plans can include volume discounts, but teams should get workload-specific pricing before comparing total costs. The $0 base fee is a useful entry point, not proof that production use will be inexpensive.
Baseten pricing FAQ
How much does Baseten cost?
Pay as you go (Basic) has a $0 monthly base platform fee, plus usage charges. Model APIs are billed per 1M tokens, but rates are not listed. Production scale, Pro and Enterprise have custom pricing, and annual pricing is not published.
Does Baseten have a free plan?
The pricing page lists a $0 base platform fee for Pay as you go (Basic) and free starter credits, but it does not publish the credit amount. Usage is billed separately, so the $0 fee does not mean workloads are free.
Does Baseten offer annual billing?
Annual pricing is not published. Pay-as-you-go usage is billed on demand, and the pricing information says volume discounts are available for custom plans. Ask Baseten for a quote if you need annual terms or committed capacity.
What extra costs should I budget for?
Usage can include Model API token charges, GPU compute, sandbox CPU and RAM, and sandbox snapshots at $0.20/GB-month. Some GPU rates are published, but Model API rates are not. Seat costs are also not published; the pricing page says there are no overage penalties.