Together AI launches post-training service for custom models
Together AI introduced Together Post-Training, a service for customizing open models with supervised fine-tuning and reinforcement learning and deploying them on the same infrastructure.
AT A GLANCE
Key points
- Supports supervised fine-tuning and reinforcement learning
- Developers can define custom losses and the full training loop
- LoRA, selected full-weight training, and dedicated GPU resources are available
- Introductory pricing is $9 per B200 GPU-hour for the first six weeks
What is changing?
Together AI introduced Together Post-Training, a service for customizing open models with proprietary data and reward signals. It combines supervised fine-tuning (SFT), reinforcement learning (RL), custom loss functions, and developer-defined training loops in one workflow.
According to the company’s official announcement, the service supports LoRA experiments, multiple adapters on the same GPU, full-weight training for selected models, and dedicated GPU resources. Asynchronous RL and generators that run during training are intended to scale reward computation and model updates. Together AI announced an introductory price of $9 per B200 GPU-hour for the first six weeks.
Supported examples include GLM 5.3, Kimi K3, and Qwen 3.8-27B. Together AI says trained models can be deployed in FP4 format. Because performance, cost, and quality depend on the use case, teams should validate results with their own datasets and evaluation criteria.
Explore the original sources
Source date: 19 October 2026.
Published by AI Radar: 19 Oct 2026, 22:05 (Istanbul, UTC+3).
FOLLOW THE TOPIC
Topics and tags
Share your discovery.
Send this story to a friend.
Reader comments
What do you think about this development? Share your experience or question. Comments are published after editorial approval.
Your name/nickname and comment become public if approved. Read: Privacy notice.
Loading comments…