350+ Tools CoveredSource Data Updated Weeklydates

Replicate Pricing: Per-Second Hardware and Per-Output Model Rates

By Egor Burlakov · Edited · Source dates below

Pricing snapshot

Scroll sideways to see all columns. With a keyboard, focus this table and use the arrow keys.

Decision pointWhat Replicate states

Billing model

You only pay for what you use. Some models are billed by hardware and time, others by input and output.

Public models

Most are billed by run time at a price per second that depends on the hardware. You pay only for the time it takes to process your request.

Private models

Most run on dedicated hardware and are billed for all the time instances are online: setup, idle and active.

Fast booting fine-tunes

Billed only while active and processing requests, with no charge for idle time. Labeled as such in the model's version list.

Enterprise and volume

Volume discounts for large amounts of spend. For complex requirements: dedicated account manager, priority support, higher GPU limits, performance SLAs, and help with onboarding, custom models and optimizations.

Which billing unit applies to you

  • Calling a public model. Check the model's page, where Replicate shows a cost estimate.

  • Deploying your own model. You package it with Cog, Replicate's open-source tool. A deployment runs on a hardware SKU from the table below.

  • Serving a fine-tune with uneven traffic. A version labeled as a fast booting fine-tune keeps quiet periods off the bill.

  • Needing H200, 2x or more H100, or 4x or more A100 or L40S. Replicate lists these under "Additional hardware" and says H200 capacity and additional multi-GPU capacity are available with committed spend contracts.

Hardware rates

Scroll sideways to see all columns. With a keyboard, focus this table and use the arrow keys.

Replicate hardware pricing, billed per second (hourly equivalent as shown by Replicate)

Hardware (SKU)Per secondPer hourGPU / CPU / GPU RAM / RAM

CPU, cpu

$0.000100/sec

$0.36/hr

  • / 4x / - / 8GB

Nvidia T4 GPU, gpu-t4

$0.000225/sec

$0.81/hr

1x / 4x / 16GB / 16GB

Nvidia L40S GPU, gpu-l40s

$0.000975/sec

$3.51/hr

1x / 10x / 48GB / 65GB

2x Nvidia L40S GPU, gpu-l40s-2x

$0.001950/sec

$7.02/hr

2x / 20x / 96GB / 144GB

Nvidia A100 (80GB) GPU, gpu-a100-large

$0.001400/sec

$5.04/hr

1x / 10x / 80GB / 144GB

2x Nvidia A100 (80GB) GPU, gpu-a100-large-2x

$0.002800/sec

$10.08/hr

2x / 20x / 160GB / 288GB

Nvidia H100 GPU, gpu-h100

$0.001525/sec

$5.49/hr

1x / 13x / 80GB / 144GB

Show full plan details

Scroll sideways to see all columns. With a keyboard, focus this table and use the arrow keys.

Additional hardware: H200 capacity and additional multi-GPU capacity are available with committed spend contracts

Hardware (SKU)Per secondPer hour

4x Nvidia A100 (80GB) GPU, gpu-a100-large-4x

$0.005600/sec

$20.16/hr

8x Nvidia A100 (80GB) GPU, gpu-a100-large-8x

$0.011200/sec

$40.32/hr

2x Nvidia H100 GPU, gpu-h100-2x

$0.003050/sec

$10.98/hr

4x Nvidia H100 GPU, gpu-h100-4x

$0.006100/sec

$21.96/hr

8x Nvidia H100 GPU, gpu-h100-8x

$0.012200/sec

$43.92/hr

Nvidia H200 GPU, gpu-h200

$0.001525/sec

$5.49/hr

2x Nvidia H200 GPU, gpu-h200-2x

$0.003050/sec

$10.98/hr

4x Nvidia H200 GPU, gpu-h200-4x

$0.006100/sec

$21.96/hr

8x Nvidia H200 GPU, gpu-h200-8x

$0.012200/sec

$43.92/hr

4x Nvidia L40S GPU, gpu-l40s-4x

$0.003900/sec

$14.04/hr

8x Nvidia L40S GPU, gpu-l40s-8x

$0.007800/sec

$28.08/hr

  • Vendor source · Observed · Stored official pricing capture for replicate. Claim-focused literal span 2000:6000 of 8055 characters; omitted text is not evidence. Review retained billing conditions against these excerpts.

How the bill is calculated

Time-billed models charge billed seconds × the hardware's per-second rate; the snapshot shows which seconds count. Models billed by input and output charge units × the model's rate. Replicate's pricing page gives these examples:

Scroll sideways to see all columns. With a keyboard, focus this table and use the arrow keys.

Example public-model rates on Replicate's pricing page

ModelPublished rate

black-forest-labs/flux-1.1-pro

$0.04 / output image

black-forest-labs/flux-dev

$0.025 / output image

black-forest-labs/flux-schnell

$3.00 / thousand output images

deepseek-ai/deepseek-r1

$3.75 / million input tokens, plus $0.01 / thousand output tokens

ideogram-ai/ideogram-v3-quality

$0.09 / output image

recraft-ai/recraft-v3

$0.04 / output image

wavespeedai/wan-2.1-i2v-480p

$0.09 / second of output video

wavespeedai/wan-2.1-i2v-720p

$0.25 / second of output video

Worked examples

Hypothetical line items at the published rates above. Each covers only the usage named, not a full bill.

  • Images: 50,000 flux-schnell output images at $3.00 per thousand output images is 50 × $3.00 = $150, for those images only.

  • Private model on one A100 (80GB): 100 online instance-hours (360,000 seconds) at $0.001400/sec is $504, for hardware time only. Each extra instance adds its own hours.

  • Text: 10 million deepseek-r1 input tokens at $3.75 per million input tokens is $37.50, for input only. Output tokens add $0.01 per thousand.

  • Video: 1,000 seconds of wan-2.1-i2v-480p output at $0.09 per second of output video is $90, for that video only.

What drives the bill

  • Hardware choice. On time-billed models, the SKU sets the per-second rate.

  • Instance count on deployments. The Node.js client sets a deployment's minimum number of instances to run and the maximum it scales up to.

  • Traffic pattern. Replicate's home page says it scales up with traffic and, with no traffic, scales down to zero and does not charge.

  • Output volume. For per-output models, image count, token count or seconds of video set the charge, not run time.

Costs that are easy to miss

  • Large input files. The Node.js client uploads files up to 100MiB. Larger inputs go in your own storage, passed as a public URL.

Our take

Replicate fits best when traffic is uneven and a public model or a fast booting fine-tune does the job, because you then pay only for processing time, or for input and output. For a steady, high-volume private model, total the hourly hardware cost first and ask Replicate about volume discounts before you commit.

Evaluate this price

Sources