Replicate pricing: per-output models and per-second hardware
Checked Nov 6, 2026 on Replicate’s official pages (sources below). Prices change often: confirm with the provider before you buy.
Replicate runs open-source and proprietary machine learning models through an API. Some public models bill by input and output (per image, per token, per second of video); most bill by the time they run on a given hardware type, priced per second.
Private models you deploy yourself with Cog run on dedicated hardware, so you pay for setup, idle and active time; fast-booting fine-tunes bill only active time. Figures below are quoted from replicate.com/pricing, checked Nov 6, 2026.
Example public models (billed by output)
- black-forest-labs/flux-1.1-pro
- Price$0.04 per output image
- black-forest-labs/flux-dev
- Price$0.025 per output image
- black-forest-labs/flux-schnell
- Price$3.00 per thousand output images
- ideogram-ai/ideogram-v3-quality
- Price$0.09 per output image
- deepseek-ai/deepseek-r1
- Price$3.75 per million input tokens; $0.01 per thousand output tokens
- wavespeedai/wan-2.1-i2v-720p
- Price$0.25 per second of output video
Hardware pricing (billed per second)
- CPU (Small)
- Per second$0.000025
- Per hour$0.09
- GPU RAM—
- CPU
- Per second$0.000100
- Per hour$0.36
- GPU RAM—
- Nvidia T4
- Per second$0.000225
- Per hour$0.81
- GPU RAM16GB
- Nvidia L40S
- Per second$0.000975
- Per hour$3.51
- GPU RAM48GB
- Nvidia A100 (80GB)
- Per second$0.001400
- Per hour$5.04
- GPU RAM80GB
- Nvidia H100
- Per second$0.001525
- Per hour$5.49
- GPU RAM80GB
- 8x Nvidia H100
- Per second$0.012200
- Per hour$43.92
- GPU RAM640GB
H200 and multi-GPU A100/H100/L40S capacity is listed with committed spend contracts. Volume discounts and enterprise support are available.
Per output or per second?
Check each model’s page: Replicate shows a cost estimate for every model. Per-output models (most image models) cost the same however long they take. Per-second models depend on the hardware and run time, so a slow model on an expensive GPU adds up. Private deployments also bill idle time unless they are fast-booting fine-tunes. Related: Together AI pricing, best AI inference APIs.
Replicate pricing FAQ
How does Replicate charge?
Per replicate.com/pricing (Nov 6, 2026): most public models bill by the seconds they run on a hardware type (for example an Nvidia L40S at $0.000975 per second), and some bill per output, such as FLUX 1.1 Pro at $0.04 per image.
What does an H100 cost on Replicate?
$0.001525 per second, which Replicate shows as $5.49 per hour; 8x H100 is $43.92 per hour.
Do private models on Replicate bill for idle time?
Yes, most private models run on dedicated hardware and bill for setup, idle and active time. Fast-booting fine-tunes are the exception and bill only while processing requests.
Sources (official pages, read Nov 6, 2026)
Spot something out of date? Email hello@youraitools.co and we’ll fix it.