From model weights to a running service
Together AI provides infrastructure for teams using open models: serverless inference APIs, dedicated model deployments, fine-tuning and GPU clusters. A developer can call a hosted model without operating its servers, or use dedicated infrastructure for a customized deployment. Model developers supply the weights; Together provides services for deploying and adapting them. [1]
The distinction matters for the AI ecosystem. Publishing weights makes a model available to download; serving requests also requires compute, deployment and capacity management. Together sells those operational services alongside tools for modifying models. Its role therefore connects model releases to the applications that use them, while the underlying model and its license remain separate choices. [1]
Adding the data-preparation layer
On May 15, 2025, Together announced its acquisition of Refuel.ai, whose models and workflow tools turn unstructured records into structured data. Together made Refuel LLM-2 available for serverless inference and LoRA fine-tuning that day. TechTarget’s contemporaneous report said the price was undisclosed. [2][3]
This added a data-processing capability beside Together’s existing model infrastructure. The announcement establishes the acquisition and the available model endpoints; broader claims about improving enterprise data quality remain the companies’ rationale for the deal, rather than a measured result of the integration. [2][3]
Sources
- Overview
Together AI · retrieved Oct 23, 2026
- Together AI acquires Refuel.ai to unlock data for developers and businesses building production-grade AI applications
Together AI · May 15, 2025
- Together AI acquires startup to improve data quality
TechTarget · May 15, 2025