en.Wedoany.com Reported - NVIDIA has launched the Nemotron series of open models, positioning them to help enterprises build controllable, trustworthy AI tailored to specific business needs through model customization, rather than simply selecting ready-made foundation models.

While enterprises currently have a wide variety of AI models to choose from, true competitiveness comes from how they leverage existing models to improve workflows, incorporate domain knowledge, and surpass standards in accuracy and trustworthiness. Open models are designed for customization, enabling enterprises to have full ownership and control, and to build specialized agents that differ from traditional AI systems.

In terms of customization, open models allow teams to test and improve based on their own data, workflows, and accuracy definitions. This is particularly important in industries such as healthcare and law, where the cost of errors is high and sensitive data must be handled. For example, NVIDIA states that with open models, teams can inspect applications, conduct private evaluations, and build reinforcement learning environments tailored to their own workflows, without routing proprietary data through third parties.

Companies across multiple industries have begun customizing Nemotron for specific domains: Abridge is customizing Nemotron to build a foundation model designed specifically for clinical conversations; Glean has built the Waldo agent search model, pairing Nemotron with a larger closed model to achieve lower latency and reduced token consumption in enterprise search scenarios; H Company has built Holotron 3 Nano by post-training Nemotron 3 Nano Omni on proprietary computer-use data, achieving over 76% accuracy on the OSWorld-Verified computer task benchmark while matching other leading frontier models at a fraction of the cost; Harvey has post-trained Nemotron 3 Ultra on legal benchmarks, achieving frontier-level accuracy on complex legal tasks with a per-run cost at least 10 times lower than leading closed models; Heidi Health produces outputs of frontier quality in clinical documentation without requiring frontier-scale computing resources; YTL AI Labs has post-trained Nemotron models for the Malay language, providing localized and customized AI capabilities to the Malaysian developer community.

Customization not only improves accuracy but also brings cost advantages. NVIDIA notes that the NVIDIA NeMo open-source library suite can accelerate model customization and evaluation. LangChain has adapted the Deep Agents framework for Nemotron 3 Ultra, achieving the highest agent accuracy among open models with a per-run cost approximately 10 times lower than leading closed alternatives. Arcee AI has post-trained Nemotron on the NVIDIA Blackwell platform, achieving an inference cost of approximately 90 cents per million output tokens—about 20 times cheaper than comparable closed frontier models—while ranking second on the PinchBench benchmark and maintaining fully open weights.

This cost advantage helps enterprises conduct broader experiments, achieve more deployments, and accelerate iteration. The NVIDIA Nemotron Coalition is driving the transformation of open model development into an ecosystem initiative, bringing together model builders and developers to improve Nemotron through shared data, evaluations, and domain knowledge. These models are available for trial at build.nvidia.com.