Post-training, Nemotron 3 Ultra reached frontier quality on Harvey's Legal Agent Bench, demonstrating a 10x reduction in cost per run. NVIDIA emphasizes that open models provide enterprises with the visibility, control, and customization needed to evaluate and specialize AI with domain knowledge.
Open-source models are closing the capability gap with proprietary models while significantly reducing inference costs for specialized enterprise AI workloads.