Enki

Replicate review: API pricing for running AI models

Run and deploy AI models through an API, billed by the second

7.8Editor’s scorePaidfrom Pay as you goPricing verified Nov 17, 2026

Screenshots

Replicate

Run AI models with one API call

7.8OUT OF 10

Enki editor score

Great

Human-vetted editorial rating

Earned by one API across image, video and language models and clear per-output prices such as $0.04 per FLUX 1.1 Pro image; cost by no published free tier, idle-time billing on private models and $5.49/hr H100s.

We may earn a commission if you sign up through our links. It never affects our ratings or which stories we cover.

Overview

Replicate hosts a large library of public models, from image generators such as FLUX to video and language models, and lets developers run any of them with a single API call. Most public models bill by the second for the GPU they run on; official models from labs such as Black Forest Labs and Ideogram bill per image, per second of video or per token instead. You can also package your own model with Replicate's open-source Cog tool and deploy it on dedicated hardware, where setup and idle time are billed too. Billing is either prepaid credit or a monthly invoice in arrears. Replicate announced on November 17, 2025 that it is joining Cloudflare and said its API is not changing.

Enki verdict

Replicate is best at putting a wide range of image, video and language models behind one simple API, with per-output prices on the official ones. Reach for it when you are adding generative media to a product and want to try models without managing GPUs, not for running a cheap high-volume text endpoint.

Key features

One API for many models

Run image, video, audio and language models with the same call.

Per-second billing

Pay only for the hardware time a public model uses.

Custom deployments

Package a model with Cog and deploy it on dedicated GPUs.

Pros & cons

What we like

  • Image, video, audio and text models behind one API
  • Per-output prices on official models are easy to budget
  • Open-source Cog for packaging your own model

Worth noting

  • No published free tier
  • Private deployments bill for idle time
  • H100 time costs more than at some GPU clouds

Replicate pricing

Replicate has no subscription: you pay for the hardware time a model uses, or per output on official models.

Is Replicate free?

No published free tier. Replicate says you can run select models for free at first, then you are asked to set up billing; it does not say how much free use that is.

  • Free use of select models before billing is required; amount not published
  • Some features need billing set up
Replicate plans and prices
Public modelsPer second of hardwareprepaid credit or monthly invoice in arrears
  • CPU $0.000100/sec ($0.36/hr)
  • Nvidia T4 $0.000225/sec ($0.81/hr); L40S $3.51/hr
  • Nvidia H100 $0.001525/sec ($5.49/hr)
Official modelsPer output
  • black-forest-labs/flux-1.1-pro: $0.04 per image
  • black-forest-labs/flux-schnell: $3.00 per thousand images
  • wavespeedai/wan-2.1-i2v-480p: $0.09 per second of video
Private modelsPer second of hardwareincludes setup and idle time on dedicated hardware
  • Deploy your own model packaged with Cog
  • Fast-booting fine-tunes bill only while active

Source: replicate.com/pricing
Checked on: . Prices can change; confirm at checkout.

What it means for you

Use official models when you want a fixed price per image or video and no surprises; per-second public models are cheaper for short jobs but harder to predict. Only move to private deployments when you have steady traffic, because dedicated hardware bills for setup and idle time as well. If you just need text models behind one key, OpenRouter is simpler, and for images without writing code, Midjourney or Stable Diffusion fit better. See the best AI coding tools.

Price history

  1. Checked against https://replicate.com/pricing: CPU $0.36/hr, T4 $0.81/hr, L40S $3.51/hr, A100 80GB $5.04/hr, H100 $5.49/hr; flux-1.1-pro $0.04 per image. No free tier on the pricing page; billing docs mention free runs of select models.

What real users think

  • “In ComfyUI using default z-image workflow, generating a single image just took me 399 seconds, … On replicate.com a single image takes 1.5s at a price of 1000 images per $1.”

  • “Replicate has had multiple ways to deploy for auto scaling … but that has always seemed like it would be too expensive for a broke bootstrapper like me so I avoided it … in general I avoided it for latency-sensitive things.”

  • “I have been using replicate.com for some fun projects. I recently encountered a problem. I submitted an official support request a week ago. There was zero response at all.”

Quoted word for word from public posts, shortened only where marked with “…”. We pick a spread of views, not the kindest ones, and these comments do not affect the Enki score. Checked Nov 18, 2026.

Replicate FAQ

Is Replicate free?
Not really. Replicate's billing docs say you can run select models for free at first, after which you are asked to set up billing, but they publish no free credit amount or free tier. Everything else is pay as you go.
How much does Replicate cost?
It depends on the model. Most public models bill by the second for their hardware, from $0.36 an hour on CPU to $5.49 an hour on an Nvidia H100. Official models bill per output, such as $0.04 per image for FLUX 1.1 Pro.
Is Replicate part of Cloudflare?
Replicate announced on November 17, 2025 that it is joining Cloudflare. It said Replicate would carry on as a distinct brand, the API would not change and existing models would keep working.
Can I deploy my own model on Replicate?
Yes. You package it with Cog, Replicate's open-source tool, and run it on dedicated hardware. Private models bill for setup and idle time as well as processing.

No written reviews yet — be the first to share your experience.

Related tools