Resources // Voice & Multimodal AI

Replicate

Offer Thousands of models, billed by the second

Suits for Non-VC-backedCreatorsResearchersOpen-source

Updated Jun 2026

Apply ↗

The offer

Run thousands of open image / video / audio / LLM models via one API, billed by the second of compute (same price for one image or a million)
Small free credit on signup to try it, then pay-as-you-go: no subscription, no minimums
GPU by the second: Nvidia T4 ~$0.81/hr, L40S ~$3.51/hr, A100 80GB ~$5.04/hr, H100 ~$5.49/hr; CPU from ~$0.09/hr
Many models billed per output instead: FLUX 1.1 Pro $0.04/image, FLUX schnell $3 / 1,000 images, Recraft V3 $0.04/image
Replicate joined Cloudflare in November 2025. Pricing is unchanged so far, but packaging may evolve

Who qualifies

Open to anyone: sign up and start. The free signup credit amount isn't published (small)
Enterprise tier (volume discounts, SLAs, higher GPU limits) via the Replicate enterprise page
No standing free tier: you pay from the first run
Private / dedicated deployments bill for setup and idle time, so a misconfigured always-on model gets expensive

Community Insights

Replicate is repeatedly cited as the low-friction way to run open models without managing GPUs. Developers like that pricing is transparent and granular (e.g. 1,000 images per $1 on cheap models). Benchmarks back its speed claims for popular models like FLUX. The headline 2025-2026 development is that Replicate joined Cloudflare, which the community is watching. The main practical caution is that there’s no free tier, so you pay from the first call and idle private models can run up cost.

Best Practices (from community tips)

Read the per-model cost estimate before running. Billing mode (per-second vs per-output) varies by model.
Prefer public models for spiky or low-volume work. They only bill active time (setup and idle are free).
Avoid private / dedicated deployments unless traffic is steady. You pay for idle and setup there.
Use cheap models for bulk jobs (FLUX schnell at $3 / 1,000 images) and cap spend on always-on models.
Write against the API rather than the web UI, so you can move providers if Cloudflare changes packaging.
For committed volume, ask Enterprise about discounts and SLAs.