How NVIDIA’s Nemotron 3 Super Targets the Cost Problem in Agentic AI

NVIDIA Corporation (NASDAQ:NVDA) is one of the best growth stocks to invest in according to billionaires.

On March 11, 2026, NVIDIA said it launched Nemotron 3 Super, an open 120-billion-parameter hybrid mixture-of-experts model with 12 billion active parameters aimed at agentic AI workloads. The company said the model is built to address two practical problems in multi-agent systems: rising token costs from long context windows and the added expense of using large models across many reasoning steps.

NVIDIA said the model has a 1-million-token context window and uses a mix of Mamba layers, transformer layers, sparse expert activation, latent MoE, and multi-token prediction. According to the company, those design choices deliver up to 5x higher throughput and up to 2x higher accuracy than the prior Nemotron Super model, while NVFP4 precision on Blackwell cuts memory use and enables inference up to 4x faster than FP8 on Hopper without sacrificing accuracy.

The company also said Nemotron 3 Super is being released with open weights under a permissive license, along with training methodology, more than 10 trillion tokens of pre- and post-training data, 15 reinforcement-learning training environments, and evaluation recipes. NVIDIA listed availability through build.nvidia.com, Perplexity, OpenRouter, Hugging Face, and several cloud and inference partners.

NVIDIA Corporation (NASDAQ:NVDA) designs graphics processors, AI computing systems, networking hardware, and software platforms for data center, gaming, automotive, and enterprise markets.

READ NEXT: 33 Stocks That Should Double in 3 Years and 15 Stocks That Will Make You Rich in 10 Years

Disclosure: None. Follow Insider Monkey on Google News.