Criterion Replicate Together AI
CategoryAI Deployment & MonitoringInference & Model Hosting
In one sentenceRun and deploy open-source and custom AI models through an API, paying per second of hardware or per output.Cloud platform for running, fine-tuning and training open models on serverless APIs or rented GPUs.
Problem it solvesUsing or deploying AI models, especially image and video models, without managing GPUs.Running open models in production without buying and operating your own GPUs.
Free planNoNo
Developer APIYes — docsYes — docs
Open sourceNoNo
PlatformsWeb, REST APIWeb, REST API
Pricing (summary)Usage-based. Public models mostly bill by hardware time (e.g. Nvidia T4 $0.81/hr, A100 80GB $5.04/hr, H100 $5.49/hr, billed per second). Some official models bill per output or token, such as flux-1.1-pro at $0.04 per image.
Checked 2026-10-19		Usage-based. Serverless models are billed per 1M tokens (for example gpt-oss-120B $0.15 / $0.60, Kimi K3 $3 / $15). Dedicated endpoints and GPU clusters are billed per GPU-hour, and fine-tuning per 1M training tokens with minimum charges.
Checked 2026-10-19
Strengths
  • Simple pay-per-use pricing
  • Huge model selection
  • Good for media models
  • Broad open-model catalogue
  • Transparent GPU-hour pricing
  • Inference, fine-tuning and training in one place
Limitations
  • No free tier mentioned on the pricing page
  • Private deployments bill idle time
  • Costs vary by model runtime and can be hard to predict
  • Free credit amount is not stated on the pricing page
  • Some hardware is contact-sales only
  • Pricing page shows two fine-tuning tables without explaining which applies