Together AI is releasing a major update to its Dedicated Model Inference platform, designed to give teams full control over open-weight model deployments without building custom infrastructure. Key features include multiple deployments behind a single stable endpoint, canary/blue-green/rolling updates with auto-rollback, A/B and shadow traffic testing, autoscaling across regions, and a Prometheus-compatible observability endpoint. The platform also introduces roughly 4× faster model warm starts via a rebuilt caching and distribution layer. Alongside this, a closed beta for custom training launches, covering full-weight and LoRA reinforcement learning and supervised fine-tuning, with checkpoints deployable directly to production inference without platform switching.
