Together AI Launches Provisioned Throughput for Reserved Inference on Open Models

Together AI announced Provisioned Throughput, providing reserved inference capacity for frontier open models like MiniMax M3 and GLM-5.2. The service uses token-based billing, offers a 99% uptime SLA, and claims costs up to 90% lower than proprietary APIs. Users need not compute GPU hours or manage infrastructure to obtain predictable inference performance. For development teams relying on open models in production, this service simplifies capacity planning and cost control. Applications requiring stable inference throughput—such as real-time chat, content generation, or large-scale batch processing—can avoid availability issues from load spikes. Token-based billing also makes costs transparent and facilitates budget management.