Together AI launches post-training service for custom models with $9 GPU-hour pricing

Experiments run on dedicated Model Resources – GPU allocations reserved for a single customer – and can host multiple training sessions such as LoRA adapters or full‑weight fine‑tuning. The SDK provides a create method for provisioning resources and per‑step calls for sampling, loss computation and weight updates. Early‑access users reported rapid setup: Adit Jain of Collinear AI highlighted seamless onboarding, Anthony Li of Deel noted the ability to start real RL experiments within hours and run five concurrent runs, and Oliver Normand of Jack & Jill said the platform matched a closed‑source model’s quality while cutting serving costs. Together’s engineers assist throughout the post‑training lifecycle, from model selection to endpoint deployment.