By Hassan Mujtaba
Publication Date: 2026-05-04 20:30:00
NVIDIA’s Open-Source “Nemotron 3 Super” AI model has topped the EnterpriseOps-Gym leaderboard, showcasing NVIDIA’s software prowess.
NVIDIA Is Topping Both AI Hardware and Software Leaderboards With Its Open-Source Nemotron 3 Super, Leading The Pack
In March this year, NVIDIA introduced its Neomtron 3 Super, a 120B AI model with 12B active parameters. Based on a hybrid MoE architecture, the model is designed to deliver a 5x throughput versus the previous Nemotron Super model, and tackles large context with a native 1M-token context windows that gives agents long-term memory for aligned, high accuracy reasoning.
Some of the highlights of NVIDIA’s Nemotron 3 Super model include:
- Latent MoE that calls 4x as many expert specialists for the same inference cost, by compressing tokens before they reach the experts.
- Multi-token prediction (MTP) that predicts multiple future tokens in one forward pass, dramatically reducing generation time for long sequences and enabling built-in speculative decoding.
- Hybrid Mamba-Transformer backbone integrating Mamba layers for sequence efficiency with Transformer layers for precision reasoning, delivering higher throughput with 4x improved memory and compute efficiency.
- Native NVFP4 pretraining optimized for NVIDIA Blackwell, significantly cutting memory requirements and speeding up inference by 4x on NVIDIA B200 compared to FP8 on NVIDIA H100, while maintaining accuracy.
- Multi-environment…
