Is Replicate free? What it actually costs in 2026

What the free tier actually includes

• —No free tier stated
• —Public models: billed by run time (some by tokens or images)
• —Private models: billed for setup, idle and active time

Where the wall is

Idle time on private deployments — you pay for the instance whether or not it is answering. Public models are pay-per-run and have no wall, only a bill.

After that: CPU (small) $0.000025 a second; Nvidia T4 $0.000225; A100 80 GB $0.001400; H100 $0.001525 — about $5.49 an hour for an H100.

Can you self-host Replicate? No — cloud only

Replicate is hosted only. The self-hosted route is the same open models on a rented GPU — the per-chip rental prices on this site show what that costs.

What it actually takes to run one yourself

Real RAM against the documented minimum, monthly cost, setup time, and the one thing that catches people out.

When paying is still the right call

Stay if your traffic is spiky, low or unproven. Scale-to-zero genuinely costs nothing between requests, and no self-hosted setup can match that — a GPU you rent is billed whether or not anyone calls it. Stay also if you are serving many different models occasionally rather than one model constantly, because that is the exact shape Replicate's economics reward. And stay while your team has no one who wants to own GPU drivers, CUDA versions and node failures at 3am. The move below is for steady, predictable load, not for experiments.

What you are locked into

Low to moderate, and lower than it looks. The weights are almost always public models you can download yourself, and Cog is open source, so a container built for Replicate runs elsewhere. What you rewrite is the calling code: predictions, webhooks and streaming are Replicate-shaped. Budget days, not weeks — and note that vLLM and TGI both expose an OpenAI-compatible API, so if your code already speaks OpenAI the swap can be a base-URL change.

If you would rather not pay: vLLM