Inside Together AI: The Research Driving Next-Gen Open-Source Models

Open-source AI models are rapidly closing the performance gap with proprietary systems, offering enterprises greater data sovereignty, customization, and cost-efficiency. Together AI facilitates this transition by providing an accelerated cloud platform that supports the entire AI lifecycle, from pre-training to inference. Key technical advancements, such as speculative decoding, enable significant inference speedups by utilizing smaller, distilled models to verify outputs from larger, more complex ones. While proprietary models once dominated, recent breakthroughs in model architectures and training strategies—such as those seen in DeepSeek’s models—demonstrate that open-source solutions can now match or exceed closed-source capabilities. As enterprises demand more control over their models and infrastructure, the industry is shifting toward flexible, open-source-centric strategies, compelling proprietary providers to adapt their business models to remain competitive in an increasingly open ecosystem.