NVIDIA's Nemotron 3.5 Super VL Vision-Language Model Leaks Ahead of GTC Announcement

NVIDIA inadvertently published documentation for its unreleased Nemotron 3.5 Super VL, a 120-billion-parameter vision-language model, before hastily removing it from public repositories. The disclosure, which surfaced through NVIDIA's official NeMo AutoModel documentation pages, revealed the architectural specifics of a hybrid Mamba-Transformer mixture-of-experts model designed to handle image, video, and text inputs simultaneously.