NVIDIA releases 120B Nemotron 3.5 Super VL and two Bolt edge models

By Splitfeed EditorialLast updated Oct 15, 2026, 17:50 UTC

On October 15, 2026, NVIDIA released three new models under the Nemotron 3.5 family. The flagship, Nemotron 3.5 Super VL, is an open-weight multimodal model with 120 billion parameters aimed at enterprise AI agents running in data centers. The two Bolt edge models, Nemotron 3.5 Bolt 74B-A9B and Nemotron 3.5 Bolt E4B, target local hardware with 24 GB and 4 GB of memory respectively. NVIDIA described all three as optimized for agentic workloads spanning data center to desktop. Adobe and NVIDIA also announced an expanded collaboration tied to agentic content creation on the same day, and dozens of organizations were reported to already be fine-tuning Super VL at launch.

The release matters beyond its parameter count because it extends capable, open-weight agentic AI to hardware most individuals and smaller organizations already own. At the same time, a separate but concurrent controversy has complicated the reception: NVIDIA's app was found to include a feature allowing AI agents to control it directly, defaulting to enabled, which critics flagged as a security and performance concern. That issue, though distinct from the model launch itself, has shaped how many people encountered the news.

The pro-AI camp greeted the models with enthusiasm. Accounts including Unsloth AI highlighted that the 74B Bolt model runs within a 24 GB memory budget while delivering strong performance for its size, and Mr. Ånand pointed to open weights as enabling custom fine-tuning for specific use cases. The anti-AI camp largely focused on the opt-out agent control feature. Theakitaneruone urged users to check their NVIDIA app settings immediately, Invertex called the default-on behavior a security risk, and Abelorion argued that running AI agents on a laptop tanks performance. The middle ground camp raised questions about independent verification: Zubiqo noted that reported wins over Claude Opus 5 and GPT-6.1 Sol on specialized benchmarks remain unverified, and Golden Era Super Intelligence observed that the model still trails top closed alternatives by a wide margin.

The near-term questions worth watching are whether independent benchmark evaluations confirm or undercut NVIDIA's performance claims, how broadly the hardware-dependency concern limits enterprise adoption outside NVIDIA infrastructure, and whether the agent-control feature controversy prompts a settings change from NVIDIA.