NVIDIA NemotronUnclaimed

NVIDIA Nemotron is a family of open models with open weights, training data, and recipes. The page describes them as delivering efficiency and accuracy for building specialized AI agents with reasoning.

The family covers several kinds of model. The reasoning models are grouped by size and purpose: Bolt runs everyday agentic tasks on laptops, Nano gives cost-efficient accuracy for specialized sub-agents, Lightning targets always-on agents, Super is for reasoning, tool calling and multimodal agentic tasks, and Ultra is aimed at the highest reasoning accuracy for complex agentic tasks. The page also lists models for visual understanding, agentic retrieval, speech and safety. Named releases include Nemotron 3.5 Super VL, Nemotron 3.5 Lightning, Nemotron 3 Ultra and Nemotron 3 Nano Omni, along with Nemotron Retrieval, Nemotron Speech and Nemotron Safety.

The training data and weights are available on Hugging Face so teams can evaluate the models before deploying them. Technical reports describing how to recreate the models are also published. The models can be run with open frameworks such as vLLM, SGLang, Ollama and llama.cpp on NVIDIA GPUs, and endpoints are offered as NVIDIA NIM microservices.

The page also covers related resources. NVIDIA publishes Nemotron datasets for pretraining, post-training, personas, safety and reinforcement learning, and says the collection includes more than 12T pretraining tokens. NeMo Switchyard is an open source model routing library that sends each step of an agent workflow to a suitable model. TensorRT-LLM is an open source inference library for Nemotron models on NVIDIA GPUs.

The audience is developers and teams building AI agents. Starter kits, cookbooks and tutorial videos cover deployment and customisation, including reinforcement learning fine-tuning in Prime Intellect Lab. The page does not state pricing for the models themselves, so none is listed here.