TL;DR: NVIDIA has released Nemotron 3 Ultra, a highly cost-effective and fast open model optimized for agentic and coding tasks.
Summary: NVIDIA has announced Nemotron 3 Ultra, an open model designed specifically for agentic harnesses and coding workflows. It is available on Hugging Face and is reportedly up to five times faster and significantly cheaper to run than comparable closed models like GPT-5.5. The model has also been integrated into Perplexity for Pro and Max subscribers.
Why it matters: AI builders can leverage a highly performant, open-weights model for long-running agents and codebase tasks at a fraction of the cost of proprietary APIs. Developers should try deploying it within their agentic frameworks or testing it via Perplexity.
Source: @NVIDIAAI