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.
Replicate offers a free option, but the vendor does not publish its limits in the published pricing details. After that, public models are charged by their usage unit, while most private models are charged for the time hardware is online. There are no monthly or annual plans listed. Your bill therefore follows the models you call, how often you call them and, for private deployments, the selected hardware and its runtime.
01Every plan, line by line
| Plan | Billed annually | Billed monthly | Built for |
|---|---|---|---|
| Public model usage | — | Quote | Teams running hosted models |
| CPU hardware | — | Quote | Private models needing CPU compute |
| A100 hardware | — | Quote | Private models needing A100 GPUs |
| H100 and H200 hardware | — | Quote | Private models needing H-series GPUs |
| L40S and one thread hardware | — | Quote | Private models needing L-series or one thread GPUs |
| Enterprise and volume | — | Quote | Organizations needing support or volume pricing |
Add-ons and overages
- Multi-GPU A100, H100, H200, and L40S capacity may require committed spend contracts; volume discounts are available for large spend.
02What the plan grid doesn't tell you
Free-tier limits
The free option's included usage limits are not published on the vendor's site. Confirm what is included before estimating a prototype's runway.
Hardware runtime
Most private models bill while hardware is online. Idle or longer-than-expected runtime can increase the bill, so account for the deployment's actual operating pattern.
Multi-GPU commitments
Multi-GPU A100, H100, H200 and L40S capacity may require committed-spend contracts. Confirm those terms before choosing a multi-GPU configuration.
Model-specific usage
Public model rates vary by output unit. Image, token and video costs differ, so a model change can alter the cost of the same application workload.
03Which plan should you buy?
| If you are… | Get | Why |
|---|---|---|
| Prototype using public models | Public model usage | Usage billing suits trying hosted models; calculate costs from each model's published unit price and your expected volume. |
| Custom model on CPU | CPU hardware | CPU Small is $0.09 per hour; CPU is $0.36 per hour. Choose based on the model's compute needs and runtime. |
| Organization needing volume pricing | Enterprise and volume | Pricing is custom, with volume discounts and listed support and onboarding options for qualifying organizations. |
04Is Replicate worth the money?
Replicate can be worth paying for when you need API access to multiple hosted models or want to deploy a custom model without running your own inference infrastructure. Usage pricing also suits workloads that are intermittent, provided you can estimate model calls or hardware runtime. The free option may be enough for an initial evaluation, but its limits are not published on the vendor's site. Teams that need predictable monthly spend, self-hosting or a contractual support commitment should confirm terms before building around the service.
Replicate pricing FAQ
Does Replicate have a free tier?
Replicate offers a free option, but the vendor does not publish its included limits in the published pricing details. Check the current pricing and account pages to understand what usage is included and when billing begins.
Are Replicate plans billed monthly?
No monthly or annual plans are listed. Public models are billed by their usage unit, and most private models are billed for the time their hardware is online. The final cost depends on model choice, usage and compute runtime.
How much does a Replicate GPU cost?
Listed rates include $5.04 per hour for one A100, $5.49 per hour for one H100 or H200, and $3.51 per hour for one L40S. Multi-GPU capacity may require a committed-spend contract.
Does Replicate offer enterprise pricing?
Yes. Enterprise and volume pricing is custom and not published. The listed offering includes volume discounts for large spend, a dedicated account manager, priority support, higher GPU limits, performance SLAs and onboarding help.