N環球 
Together AI Launches Post-Training Service for Custom Models with RL and Fine-Tuning
Together AI on October 19 launched Together Post-Training, a managed service letting developers build custom models with their own reinforcement learning and supervised fine-tuning loops while Together runs the GPU infrastructure.
Why it matters
Together AI on October 19 officially launched Together Post-Training, turning model post-training — previously the domain of well-resourced labs — into a managed service. Developers define their own training loop, including data, environments and agentic harnesses, while Together co-optimizes the underlying GPU systems, aiming to lower the infrastructure barrier around reinforcement learning (RL).
The service launches with open-weight models including Muse Spark Open, GLM 5.3, Kimi K3 and Qwen 3.8-27B, with more models to be added in the coming weeks. Introductory pricing is $9 per GPU-hour of B200 usage for the first eight weeks.
Full report
Together says the service supports LoRA training, multiple LoRA adapters sharing the same GPU capacity, full-weight experiments for selected models, asynchronous RL and custom training objectives. Trained checkpoints can be deployed natively to inference, with support for serving in FP4 precision.
The launch note cites early-access customers Deel, Jack & Jill and Collinear AI, which described faster experiment setup and the ability to run RL workloads on shared GPU capacity (company-attributed, not independently verified).
For context, Together announced an $800 million Series C in July along with commitments for more than 500 MW of independently capitalized compute capacity, and on September 29 announced with Mubadala it was exploring UAE AI infrastructure opportunities and an Abu Dhabi presence. The launch is a step toward converting its compute footprint into product revenue.
💬 Comments
No comments yet. Be the first!