Replicate

Acquired by Cloudflare — verified 2026-09-30

Cloudflare announced Replicate joining it on 2025-11-17; the audit’s verified facts record the deal as closed 2025-12-01. Cloudflare says existing Replicate APIs and workflows continue uninterrupted and Replicate stays the community hub. Planned: bring the 50,000+ Replicate models to Workers AI, add fine-tuning, and use Cog for custom models on Cloudflare’s network.

Replicate

by cloudflare (formerly independent)

Cloud platform for running and deploying AI models via API, with per-second billing and automatic scaling.

See https://replicate.com

Features

• Large catalog of community and open-source models (Cloudflare cites 50,000+), run through one API.
• Custom models packaged with Cog, Replicate’s open-source containerization tool (cog predict, cog push).
• Fast-booting fine-tunes that bill only for active processing time; private models run on dedicated hardware and bill for setup, idle and active time.
• Streaming output, webhooks, Python and JavaScript clients, web UI to try models.
• Enterprise options: volume discounts, dedicated accounts, performance SLAs.

Billing (replicate.com/pricing, 2026-09-30)

Pay-as-you-go. Most public models bill by runtime (per second of hardware); some (for example image generation) bill per input/output. Hardware options range from small CPU through Nvidia T4, L40S, A100 and H100 to 2x A100.

Multi-GPU configurations up to 8x H100 need committed contracts. No free tier is mentioned on the pricing page. Current prices: see the vendor pricing page https://replicate.com/pricing

Python example

import replicate  
  
output = replicate.run(  
    "owner/model-name",          # pick a model from replicate.com/explore  
    input={"prompt": "a painting of a cat"}  
)  

Cog workflow

1. Write Python inference code and a cog.yaml declaring the environment and input/output schema.
2. Test locally with cog predict.
3. Publish with cog push; the model gets an API endpoint.

Good fit

Teams without MLOps infrastructure, bursty workloads, experimentation across many open models. Alternatives: fal (fal-dot-ai), Kie.ai (kie-dot-ai), self-hosting. See ai-api-platforms.

Removed from earlier version

Healthcare-imaging examples and a claim of “scale to millions of users” were illustrative and unsourced.

Related

cloudflare, fal-dot-ai, kie-dot-ai, ai-api-platforms, daytona

Sources

• Cloudflare blog, Replicate is joining Cloudflare, https://blog.cloudflare.com/replicate-joins-cloudflare/ (accessed 2026-09-30)
• Replicate pricing, https://replicate.com/pricing (accessed 2026-09-30)
• Closing date 2025-12-01: KB-AI-AUDIT-2026-09 Appendix A (secondary sources; not re-checked on a primary page)