Slug: replicate
Date: 2026-09-21 (America/Chicago)
Website: https://replicate.com
Industry: AI Model Hosting / Inference Marketplace
Replicate is a developer platform to run, fine-tune, and deploy machine-learning models via API (and web UI) without managing GPUs. Community and official models are runnable with one line of code; custom models package with open-source Cog and deploy on Replicate’s cluster (FACT — homepage / pricing).
Company blog (2025-11-17) and Cloudflare posts: Replicate is joining Cloudflare as a distinct brand; API and models continue; catalog (company claim 50,000+ models) to integrate with Workers AI (FACT — replicate.com/blog/replicate-cloudflare; Cloudflare blog Dec 1, 2025 “officially part of Cloudflare”). We do not invent purchase price or private ARR.
Teaching arc: Public pay-per-prediction (GPU-sec or per-output) → Fine-tune training meters → Private models (setup/idle/active) → Deployments (scaled dedicated endpoints) → Multi-GPU / H200 committed spend → Enterprise volume & SLAs.
# Replicate — research brief **Slug:** replicate **Date:** 2026-09-21 (America/Chicago) **Website:** https://replicate.com **Industry:** AI Model Hosting / Inference Marketplace ## Label key - **FACT** — primary: replicate.com pricing, docs/billing, enterprise, company blog - **SECONDARY** — press / third-party (cited, not invented) - **INFERENCE** — reasoned from public product shape; not a private metric ## Company snapshot Replicate is a developer platform to **run, fine-tune, and deploy** machine-learning models via API (and web UI) without managing GPUs. Community and official models are runnable with one line of code; custom models package with open-source **Cog** and deploy on Replicate’s cluster (FACT — homepage / pricing). Company blog (2025-11-17) and Cloudflare posts: Replicate is **joining Cloudflare** as a distinct brand; API and models continue; catalog (company claim **50,000+** models) to integrate with Workers AI (FACT — replicate.com/blog/replicate-cloudflare; Cloudflare blog Dec 1, 2025 “officially part of Cloudflare”). We do not invent purchase price or private ARR. ## Revenue / pricing ladder (FACT — replicate.com/pricing + docs/topics/billing) 1. **Public model predictions (shared pool)** — pay only for **active processing** time; setup and idle are free to the customer. Most models bill by hardware second; some official/proprietary models bill by **output image**, **video second**, or **tokens** (FACT pricing examples). 2. **Hardware meter (GPU/CPU seconds)** — published rates include e.g. CPU Small **$0.000025**/sec (**$0.09**/hr); CPU **$0.000100**/sec (**$0.36**/hr); T4 **$0.000225**/sec (**$0.81**/hr); L40S **$0.000975**/sec (**$3.51**/hr); A100 80GB **$0.001400**/sec (**$5.04**/hr); H100 **$0.001525**/sec (**$5.49**/hr); multi-GPU and H200 tiers higher / often **committed-spend** (FACT pricing). 3. **Official / proprietary output meters** — examples on pricing page: FLUX 1.1 Pro **$0.04**/image; FLUX Dev **$0.025**/image; FLUX Schnell **$3.00**/thousand images; Ideogram v3 Quality **$0.09**/image; Recraft V3 **$0.04**/image; Wan 2.1 I2V 480p **$0.09**/sec video; 720p **$0.25**/sec (FACT pricing). 4. **Fine-tuning / training** — billed as training runtime on selected hardware (e.g. FLUX fine-tune docs: 8×H100 at **$0.012200**/sec ≈ **~$1.46** for a ~2-minute typical run) (FACT docs get-started/fine-tune-with-flux). 5. **Private models** — dedicated hardware; bill **setup + idle + active** (exception: **fast-booting fine-tunes** bill active-only like public) (FACT billing docs). 6. **Deployments** — control hardware and min/max instances; bill setup + idle + active for online instances; used for stable endpoints, avoiding shared-queue cold boots (FACT billing). 7. **Enterprise / volume** — dedicated AM, priority support, higher GPU limits, performance SLAs, onboarding/custom-model help, volume discounts, reserve compute; **Talk to us** (FACT enterprise + pricing). Teaching arc: **Public pay-per-prediction (GPU-sec or per-output) → Fine-tune training meters → Private models (setup/idle/active) → Deployments (scaled dedicated endpoints) → Multi-GPU / H200 committed spend → Enterprise volume & SLAs**. ## Value proposition themes (FACT product) - Thousands / 50,000+ production-ready models via one API (company claim — enterprise/homepage) - Pay only while code runs (public / fast-boot); scale to zero when idle (homepage) - Cog packages custom models → auto API + cloud scale - Fine-tune (e.g. FLUX) without owning GPUs ## Partners / ecosystem (FACT / company-stated) - **Cloudflare** — parent / joining narrative since Nov–Dec 2025 (company + Cloudflare blogs) - **Model labs / publishers** on catalog: Black Forest Labs (FLUX), Ideogram, Recraft, DeepSeek, Wavespeed/Wan, Meta (Llama examples in docs), Stability (docs examples), community creators (FACT pricing + docs) - **Cog** open-source packaging (FACT homepage) - Named “trusted by” logos appear on enterprise page (company marketing; treat as FACT claim, not audited customer list) ## What we do NOT invent - No ARR, burn, headcount, prediction volume, or acquisition purchase price as FACT metrics - Cloudflare deal economics labeled only as company/Cloudflare announcements, not modeled here ## Primary sources - https://replicate.com/pricing - https://replicate.com/docs/topics/billing - https://replicate.com/enterprise - https://replicate.com/home - https://replicate.com/docs/get-started/fine-tune-with-flux - https://replicate.com/blog/replicate-cloudflare - https://blog.cloudflare.com/why-replicate-joining-cloudflare/