NVIDIA company blog Oct 15: the firm released Nemotron 3.5 Super VL, an open 1… · @AIInvestHK · TwKit

@AIInvestHK

NVIDIA company blog Oct 15: the firm released Nemotron 3.5 Super VL, an open 120-billion-parameter multimodal model for always-on agents, plus two edge builds branded Bolt. Super VL and Bolt 74B-A9B 24GB are live now. Bolt E4B 4GB is dated Oct 23. A same-day NVIDIA developer post puts the Super VL MoE at 12 billion active parameters per token. The main blog says it targets roughly one-fifth the size of the largest Nvidia reasoning model and claims up to 5x faster throughput versus other open models in its class on the Artificial Analysis board it cites. That speed line is from the company. Weights are on Hugging Face and as NVIDIA NIM. Clock starts on the Oct 23 Bolt E4B date. After that, the tell for $NVDA is whether the enterprise post-training names on the launch blog turn into live agents, not just logos.

曝光 33 · 评论 0 · 点赞 0 · 书签 0