NVIDIA Nemotron 3 Ultra: Open Model for Long AI Agents

NVIDIA has released Nemotron 3 Ultra, a new open-weight model designed for long-running AI agents. This model focuses on multi-step processes like planning, tool calling, and code execution, rather than just single prompts. Here's the thing: most models are judged on quick benchmark scores. But Nemotron 3 Ultra tackles the real costs of agentic AI, where expenses and bottlenecks grow with each step of a long task. What's interesting is its key advantages. It offers up to five times faster inference and up to 30 percent lower cost per completed task on long agent chains. It also maintains coherence and efficiency over extended sessions. The model is strong in complex planning, tool use, code generation, and long-document analysis. Being open-weight means teams can fine-tune it and deploy it in their own data centers, maintaining full control. The bottom line: NVIDIA suggests evaluating agent models by the cost and time to complete an entire task, not just single-prompt performance, which could significantly impact AI deployment.