whichdevtool?
Menu
StacksDirectory Pricing changes Glossary
VPS and cloud servers App hosting and PaaS Managed databases Transactional email APIs Email marketing and newsletters Domain registrars CDN, DNS and edge security Monitoring, uptime and error tracking Auth and identity Frontend and full-stack frameworks Backend frameworks UI component libraries AI model APIs File and object storage Payments and billing Background jobs, queues and workflows ORMs and database toolkits

Model hosting API (image, video, audio and LLMs)

Replicate

API for running community and official AI models, mostly image, video and audio, billed per second of hardware or per output.

Visit replicate.com ↗ Alternatives Trial
Replicate homepage
§1

Overview

Replicate is an API for running AI models that other people have published, plus your own. The catalogue leans toward generative media: image models such as FLUX, Nano Banana, gpt-image and Seedream, video models such as Wan, and speech and music models, alongside some LLMs. Each model page shows an estimated cost per run.

There are two billing styles. Most public models charge for the seconds of hardware they use, from $0.81 an hour on a T4 to $5.49 an hour on an H100. Official models, maintained with their creators, charge per output instead: $0.04 per FLUX Pro image, or $3.75 per million input tokens for DeepSeek R1. Private deployments of your own models bill for all uptime, including idle time, unless they are fast-booting fine-tunes.

You can try select models free before Replicate asks for billing details, then choose prepaid credit or monthly billing in arrears. Replicate may charge early the first time usage crosses certain thresholds, as a fraud check.

§2

Pricing and plans

Plan Price What you get
CPU (small) $0.09/hr $0.000025 per second
Nvidia T4 $0.81/hr $0.000225 per second
Nvidia L40S $3.51/hr $0.000975 per second
Nvidia A100 (80 GB) $5.04/hr $0.0014 per second
Nvidia H100 $5.49/hr $0.001525 per second
FLUX Pro (official) $0.04 per image FLUX Schnell $3 per 1,000 images
DeepSeek R1 (official) $3.75 in / $10 out per 1M Output listed as $0.01 per 1,000 tokens

Read 2026-09-29 on replicate.com/pricing and the billing docs. Public community models bill by run time at the hardware rate; official models bill per output (image, video second or token). Private deployments bill for all uptime including setup and idle, except fast-booting fine-tunes, which bill only while processing. You can run select models for free before being asked to set up billing; the size of that allowance is not stated. Billing is prepaid credit or monthly in arrears. No spend-limit feature was described in the pages read.

Source: replicate.com/pricing · prices as published, excluding tax.

§3

What it does

Large model catalogue
Public image, video, audio and language models callable with one API.
Official models
Maintained models billed per image, per video second or per token instead of per second of GPU.
Custom deployments
Deploy your own or fine-tuned models; fast-booting fine-tunes bill only while running.
Per-second billing
Public models charge by hardware time, from $0.000025 per second on small CPUs.
Prepaid or arrears billing
Buy credit up front or pay monthly for the previous month's usage.
§4

Strengths and limitations

Strengths

  • Broad choice of image, video and audio models behind one API.
  • Per-output pricing on official models makes costs easy to estimate.
  • No GPU management for running open models.

Limitations

  • Per-second billing on community models makes cost depend on model speed and cold starts.
  • Private deployments bill for idle time as well as active time.
  • The free allowance is small and not quantified.
  • LLM prices can be higher than specialist token providers (DeepSeek R1 at $10 per 1M output tokens).
§5

Who it suits

Good for

  • Apps that generate images, video or audio
  • Trying many community models quickly
  • Serving a fine-tuned image model without running GPUs

Look elsewhere if

  • High-volume text generation, where token-priced LLM providers cost less
  • Always-on private deployments on a tight budget
§6

Alternatives to Replicate

Tool From Free option Stages
GroqCloud Inference provider (open-weight models)
Usage-based Free tier
Gemini API AI model API (model maker)
Usage-based Free tier
Mistral API (Mistral Studio) AI model API (model maker)
Usage-based Free tier
Scaleway Generative APIs Inference provider (EU-hosted open-weight models)
Usage-based Trial
OpenRouter Model router (unified API)
— Free tier
DeepInfra Inference provider (open-weight models)
Usage-based Paid

All AI model APIs →

§7

Questions

Is Replicate free?

You can run select models free for a while, after which Replicate asks you to set up billing. There is no ongoing free tier.

How does Replicate charge for models?

Most public models are billed by the second of hardware time; official models are billed per output such as per image or per token.

How much does an H100 cost on Replicate?

$0.001525 per second, about $5.49 per hour, on 2026-09-29.

Sources read for this page