What is Replicate?
Run open models or deploy your own behind one API.
Who is Replicate best for?
Replicate is best for developers who want to run and fine-tune AI models through one API.
Apps that call many models
Thousands of community and proprietary models for images, video, audio and text run behind one API with a line of code.
Fine-tuning on your own data
Train an image model on a person, object or style, and the result becomes a new model you can call like any other.
Deploying custom models
Package a model with Cog, Replicate's open-source tool, and it scales up and down with traffic on the hardware you pick.
Spiky or low-volume workloads
Public models bill only for the seconds spent on your request, so there is no idle GPU to pay for between calls.
How much does Replicate cost?
Replicate has no subscription: you pay per use, either per second of hardware from $0.09 an hour on a small CPU to $5.49 an hour on an Nvidia H100, or per output on some models, such as $0.04 per FLUX 1.1 Pro image.
- $0.000025 a second
- 1 CPU and 2GB RAM
- $0.000225 a second
- 16GB GPU RAM, 4 CPUs, 16GB RAM
- $0.000975 a second
- 48GB GPU RAM, 10 CPUs, 65GB RAM
- $0.001525 a second
- 80GB GPU RAM, 13 CPUs, 144GB RAM
Public models bill only for the time spent on your request; private models and deployments also bill for setup and idle time, except fast booting fine-tunes. Some models charge per output instead: FLUX 1.1 Pro is $0.04 an image, FLUX Schnell $3 per thousand images, Wan 2.1 image-to-video $0.09 a second at 480p or $0.25 at 720p, and DeepSeek R1 $3.75 per million input tokens and $0.01 per thousand output tokens. Each model page shows its own cost estimate. An Nvidia A100 (80GB) is $5.04 an hour, and multi-GPU setups run up to $43.92 an hour for 8x H100 or 8x H200, with larger capacity on committed spend contracts. Failed runs on public models are not charged, and Enterprise customers get volume discounts. Full pricing on replicate.com. We checked these prices on .
How do readers rate Replicate?
No one has rated Replicate yet. Be the first.
Sign in to rate. One rating per account.
How do I connect Replicate to my AI agent?
Replicate MCP is Replicate's official MCP server. Add it to your agent once, then ask it to run models hosted on Replicate to create and process media.
Gives your agent access to the models on Replicate, so it can find one for a job, run it and return images, video, audio or other output.
$ claude mcp add --transport http replicate https://mcp.replicate.com/sse
Claude Code connects to Replicate over the web.
Replicate's setup guide covers the sign-in.
Then ask your agent
- Runs
- On the web
- Sign-in
- See guide
- Maker
- Replicate
- Guide
- mcp.replicate.com
What are the best Replicate alternatives?
One tool on Recaply does Replicate's job. It is open source, so you can run it on your own server.
Questions about Replicate
What is Replicate?
Run open models or deploy your own behind one API.
Who is Replicate for?
Replicate is best for developers who want to run and fine-tune AI models through one API.
Is Replicate free?
Only at first. Replicate lets you run select models for free, but after a while it asks you to set up billing. From then on you pay per use, by the second of hardware time or per output, with no monthly fee.
How much does Replicate cost?
Replicate's paid plans are CPU (Small) at $0.09 per hour, billed by the second, Nvidia T4 GPU at $0.81 per hour, billed by the second, Nvidia L40S GPU at $3.51 per hour, billed by the second and Nvidia H100 GPU at $5.49 per hour, billed by the second.
Does Replicate have an MCP server?
Yes. Replicate MCP is Replicate's official MCP server, so agents such as Claude Code, Cursor, Codex and GitHub Copilot can run models hosted on Replicate to create and process media. Replicate runs it on the web.
Is there an open source alternative to Replicate?
Yes. Modular is an open source tool that does the same job, so you can run your own.
What are the best Replicate alternatives?
The closest alternatives on Recaply are Modular.
Where can I try Replicate?
On its official website, replicate.com.
Do I pay for idle time on Replicate?
Not on public models: Replicate bills only the seconds a public model spends on your request. Private models and deployments run on dedicated hardware, so you also pay for setup and idle time, except fast booting fine-tunes, which bill active time only.
Anaconda
Epigos AI
Qualcomm AI Hub
VModel
Prime Intellect
ZaneOps