Replicate
Run, fine-tune and deploy AI models with one API call
Overview
Replicate runs thousands of image, video, audio and language models through a single API, and lets developers package and deploy their own models with the open-source Cog tool. Models are billed per output or per second of hardware. Cloudflare completed its acquisition of Replicate on 1 December 2025; the brand and API continue.
Pricing
Pricing shown for reference only. These figures reflect RECATOOLS research as of 14 Oct 2026 and may be out of date or incomplete. This is not financial or purchasing advice — always confirm the current price on the provider’s official website before making any decision.
- FLUX schnell $3 per thousand images
- FLUX dev $0.025 and FLUX 1.1 Pro $0.04 an image
- DeepSeek-R1 $3.75 per million input tokens
- Wan 2.1 video $0.09 to $0.25 per second of output
- Rate depends on the hardware the model uses
- Billed for setup, idle and active time
- Fast-booting fine-tunes billed only while active
- Automatic scaling up and down
- 2GB RAM on cpu-small
- 8GB RAM on cpu
- Multi-GPU A100, H100, H200 and L40S on committed spend contracts
- Dedicated account manager
- Priority support
- Higher GPU limits
- Performance SLAs
Use cases
What you can produce with Replicate
- Images, video, speech and music generated by calling hosted models through one API
- Open-weight language models such as DeepSeek-R1 run without provisioning GPUs
- Image models fine-tuned on your own subject or style and served through the same API
- Your own model packaged with Cog and deployed on dedicated, autoscaling hardware
- Side-by-side model comparison in the browser playground before writing code
- Model discovery and runs from Claude, Cursor and VS Code via Replicate's MCP server and agent skills
Replicate is still one of the quickest ways to try a new model and ship it: one API, official models from Google, OpenAI, Black Forest Labs and ByteDance alongside community uploads, and Cog for publishing your own. Prices are published per output or per second of named hardware, so costs can be estimated before you build.
Watch dedicated deployments, which bill while idle, and cold starts on long-tail models. Cloudflare now owns it and has promised the API stays the same; check how the Workers AI integration develops before betting a roadmap on either side. Strong for prototyping and mixed-model products.
What we found
Replicate is a platform for running machine-learning models through an API, started in San Francisco in 2019 by a team including Ben Firshman. Its first tool was Cog, an open-source format for packaging a model; Replicate then became the place to share Cog models and call them as cloud endpoints. When Stable Diffusion arrived in 2022 it was ready for the demand, and it has stayed a quick way to put a new image, video, audio or language model behind one API call without handling GPUs.
Ownership changed late in 2025. Cloudflare announced the deal on 17 November 2025, and its SEC filings record that it bought all of Replicate, Inc. on 1 December 2025 for $57.4 million in cash consideration. Firshman's note to users said Replicate would remain a distinct brand, the API would not change and existing models would keep working. Cloudflare says it plans to bring the catalogue, which it put at more than 50,000 models and fine-tunes, to its Workers AI service. The site now runs under Replicate, LLC, and job openings are on Cloudflare's careers page.
Billing comes in two shapes. Many public models are charged by output: FLUX dev at $0.025 an image, FLUX schnell at $3 per thousand, DeepSeek-R1 at $3.75 per million input tokens, Wan 2.1 video at $0.09 to $0.25 per second. Others, and most private models, are charged per second of hardware: an Nvidia H100 at $0.001525 a second ($5.49 an hour), an A100 80GB at $5.04 an hour, a small CPU at $0.09 an hour. Private models on dedicated hardware bill for setup and idle time too, except fast-booting fine-tunes.
That idle billing, and cold starts on less-used community models, are the costs to watch at scale. For teams that want breadth and speed of integration more than the lowest price per token, it remains a strong default.
Compiled by RECATOOLS from public sources: vendor documentation and pricing, code repositories and published reporting.
About this listing
This entry was compiled from publicly available data including Replicate's official website, press releases, documentation, and reputable third-party publications. RECATOOLS is not affiliated with Replicate unless explicitly stated.
Third-party AI tools update their pricing, features, availability, and policies frequently. Information here may be outdated by the time you read this — we make reasonable efforts to keep listings current, but cannot guarantee absolute accuracy.
For the latest details, please refer to Replicate directly →
Spotted something out of date? Suggest an update →
Related reading
Alternatives to Replicate
More in Agents & Automation
