Overview
- NVIDIA released Nemotron 3.5 Lightning on Tuesday as an open 30‑billion‑parameter Mixture‑of‑Experts (MoE) model that the company says boosts output up to four times and cuts multi‑agent task completion by about 30 percent.
- The company published the model and weights on public hubs including Hugging Face, ModelScope, OpenRouter and NVIDIA’s site so developers can run it on RTX PCs, Jetson devices, DGX systems and cloud instances.
- Reporting from The Information, cited by participants, says NVIDIA is developing Nemotron 4 with plans for variants that could exceed one trillion parameters but training is incomplete and no release date has been confirmed.
- NVIDIA has joined other firms in an industry effort to build and share AI and cybersecurity tools and has publicly backed open weights as part of its strategy to broaden access to advanced models.
- The move pairs lower‑cost open models and MoE efficiency with wider device compatibility, a shift that could increase demand for NVIDIA hardware while raising ongoing questions about governance and misuse of open weights.