Replicate pricing, price hikes and cheaper alternatives
Replicate (replicate.com) is a GPU cloud platform. Pay-as-you-go hardware time billed per second (private deployments also pay for idle time), with some public models billed per output or per input/output unit instead.. No free plan.
What Replicate costs
- 100 H100 hours: $549/mo (1x Nvidia H100 GPU at $5.49/hr (CPU and RAM bundled: 13x CPU, 144GB RAM). 100 hours x $5.49 = $549. No plan or platform fee on the page. Region not stated.)
- 24/7 GPU endpoint: $2,562.30/mo (Cheapest listed GPU with at least 24GB VRAM is 1x Nvidia L40S (48GB) at $3.51/hr; the T4 has 16GB, and L4 and A10G are not listed. 730 hours x $3.51 = $2,562.30. CPU and RAM are bundled, and no idle rate is published, so the active rate applies. The page does not say which SKUs deployments can use or in which region, so L40S availability there is assumed.)
- 1,000 sandbox hours: not offered (No sandbox product appears on the pricing page, so not offered.)
Plan compared: Pay-as-you-go, 1x Nvidia H100 GPU at $5.49/hr. Prices are USD per month at on-demand, non-interruptible rates for three fixed workloads: 100 hours on one H100 80 GB GPU, one 24 GB-class GPU serving a model 24/7 (730 hours), and 1,000 hours of 2 vCPU / 4 GB agent sandboxes. They include required plan fees and separately billed CPU and RAM, net of recurring monthly credits; a dash means the product does not do that job.
What Replicate does
Core: Dedicated model endpoints, Ready-to-run model APIs, Public model and app hub, Open-weight models, Image and video generation.
Important: Serverless GPU functions, Managed fine-tuning, Bring your own container (partly), API, CLI and SDKs, Multi-lab model catalog (partly).
Also has: H200 and Blackwell GPUs (partly), Multi-GPU nodes and clusters (partly), Batch jobs and queues (partly), Spot and reserved discounts (partly), Open source and own cloud (partly), Image and PDF input (partly), Voice and realtime audio (partly), Embeddings and reranking (partly).
Replicate price hikes
No price hikes on record for Replicate.
Cheaper alternatives to Replicate
At 100 H100 hours a month, Replicate costs $549/mo. 3 GPU cloud platforms cost less:
- Runpod: $199/mo (Pay-as-you-go Pods, Community Cloud: 1x H100 PCIe 80 GB at $1.99/hr for 100 hours a month, with no plan fee or credit.), covers 89% of Replicate; missing or partial: Public model and app hub (partly), Managed fine-tuning (partly)
- Modal: $434.93/mo (Starter (free plan): $0 fee, $30/mo compute credit applied, base (no-premium) rates), covers 70% of Replicate; missing or partial: Ready-to-run model APIs, Public model and app hub (partly), Managed fine-tuning (partly), API, CLI and SDKs (partly)
- Hugging Face: $450/mo (Free account, no subscription; 1x H100 80GB on Inference Endpoints, billed hourly.), covers 91% of Replicate; missing or partial: Serverless GPU functions (partly), Managed fine-tuning (partly), Spot and reserved discounts
Replicate alternatives: 3 cheaper GPU cloud platforms compared, at every size.
Other kinds of products that cover part of Replicate
- Cloudflare Workers (Hosting): covers 55% of Replicate; missing or partial: Dedicated model endpoints (partly), Public model and app hub (partly), Serverless GPU functions, Managed fine-tuning, H200 and Blackwell GPUs
- Vercel (Hosting): covers 55% of Replicate; missing or partial: Dedicated model endpoints (partly), Public model and app hub (partly), Serverless GPU functions, Managed fine-tuning, H200 and Blackwell GPUs
- DigitalOcean Apps (Hosting): covers 34% of Replicate; missing or partial: Dedicated model endpoints (partly), Ready-to-run model APIs, Public model and app hub, Serverless GPU functions, Managed fine-tuning
Sources
Pricing page: replicate.com/pricing. Prices last checked Oct 16, 2026. Prices from replicate.com/pricing; two separate extractions of the page agreed on every SKU. Change history is unchecked: PricingSaaS has no page for this product (404) and the web search budget ran out, so changes is empty.