Replicate Pricing

A typical SDXL image on Replicate costs about $0.012 per prediction. FLUX models range from $0.003 to $0.04 per image depending on resolution and model variant. Costs are billed per second of GPU time.

Replicate offers a limited number of free predictions for new accounts. After the free allowance, all usage is billed per second of compute time with no monthly subscription required.

Each model runs on different GPU hardware. Lightweight models use T4 GPUs at $0.000225/s while large models require A100 or H100 GPUs at $0.001050/s to $0.001525/s, so the per-prediction cost depends on hardware and inference time.

For low or variable workloads, Replicate is cheaper because you avoid idle GPU costs. For sustained high-volume generation above 50,000 images per month, a dedicated GPU instance may be more cost-effective.

Wireflow charges a flat rate per generated output instead of per GPU second. This makes costs predictable and eliminates billing surprises from cold starts, queue delays, or variable inference times.

Yes, cold-start time counts toward billed seconds on Replicate. Public models can take 5 to 30 seconds to spin up, and that startup time is included in your bill for the first request.

Replicate enterprise plans include volume discounts, dedicated GPU allocation, priority support, and higher concurrency limits. Pricing is custom and requires contacting their sales team directly.

Replicate allows setting a monthly spending limit in the dashboard. Once reached, API calls are rejected until the next billing cycle. Wireflow offers per-project and per-workflow limits for more granular control.