Together AI

Running open source models usually means renting GPUs, wiring up servers and watching the bill. Together AI is an AI cloud and API platform that hosts models for inference, fine-tuning and training, so you can call them by token or rent the hardware directly.

Serverless inference is priced per million tokens, with model prices from $0.06 to $7.00, and batch inference carries an introductory 50% discount.

Dedicated Endpoints let you deploy a model on your own GPU with per-minute billing.

Options include RTX-6000, L40, A100, H100 and H200 cards, from $1.49 to $4.99 per hour.

Fine-tuning covers supervised LoRA, full fine-tuning and DPO, billed per million tokens processed.

For larger jobs, Together GPU Clusters start at $1.30 per hour for A100s, with H100, H200 and Blackwell hardware also offered.

Together AI also sells a code sandbox and a code interpreter, priced per session.

The pricing is varied, so check the model you plan to use.