Together AI Expands Inference Platform with Custom Model Hosting

Together AI has announced a major expansion of its inference platform, adding custom model hosting, dedicated GPU clusters, and a managed fine-tuning pipeline. The company is positioning itself as a comprehensive AI infrastructure provider that bridges the gap between managed APIs and raw cloud GPU access.

Custom model hosting allows enterprises to deploy their proprietary or fine-tuned models on Together's optimized inference infrastructure. The platform handles scaling, load balancing, and optimization automatically, achieving inference speeds that are typically 2-3x faster than self-hosted deployments on equivalent hardware.

Dedicated GPU clusters give enterprise customers reserved compute capacity with guaranteed availability, rather than sharing infrastructure with other users. Clusters can be configured with a mix of GPU types optimized for different workloads, from small real-time models to large batch processing jobs.

The managed fine-tuning pipeline enables customers to fine-tune open-source models on their own data using Together's infrastructure. The platform supports LoRA, QLoRA, and full fine-tuning, with built-in evaluation and automatic deployment of the fine-tuned model to production inference endpoints.

Together AI has raised over $600 million in total funding and reports serving over 500 enterprise customers. CEO Vipul Ved Prakash noted that the expansion reflects growing enterprise demand for infrastructure that offers the flexibility of open-source models with the convenience of managed services.