Realtime AI News
NVIDIA Releases Nemotron 3.5 Super VL and Two Bolt Edge Models
NVIDIA has released Nemotron 3.5 Super VL, an open 120-billion-parameter model that it says delivers the fastest task completion and the highest accuracy for specialized always-on agents. Two edge models joined it: Nemotron 3.5 Bolt 74B-A9B 24GB, tuned to balance intelligence and speed per GB for local agentic workloads, and Nemotron 3.5 Bolt E4B 4GB.
NVIDIA has expanded its Nemotron family with the launch of Nemotron 3.5, a generation that reaches from data-center deployments all the way down to desktop and edge devices. The release is led by Nemotron 3.5 Super VL, a 120-billion-parameter open model, alongside two new Bolt models aimed at edge hardware.
According to NVIDIA's blog, Super VL is built for specialized always-on agents, the class of applications that stay online and keep responding to tasks over long periods. The company describes it as delivering the fastest task completion and the highest accuracy in that role.
The two edge releases are Nemotron 3.5 Bolt 74B-A9B 24GB and Nemotron 3.5 Bolt E4B 4GB. NVIDIA positions the 74B-A9B version around a balance of intelligence and speed per gigabyte for local agentic workloads, while the E4B variant pushes the footprint into the 4GB class.
The framing is as notable as the models. By naming a per-gigabyte efficiency metric for the 74B-A9B version, NVIDIA is speaking directly to teams that must decide what an agent can run on a laptop, a workstation or a small local server rather than a rented cluster.
Chinese-language coverage echoed the same emphasis. A cnBeta report circulated through Google News framed the launch as a 120-billion-parameter flagship arriving together with lightweight edge AI models, matching the data-center-to-desktop line NVIDIA itself used.
The reason this generation matters is that always-on agents have a different cost profile from chat assistants. A model that is called continuously, all day, is highly sensitive to inference speed and cost per call, and whether it fits inside 24GB or something smaller often decides whether it can be deployed at all.
Three things are worth watching. Independent benchmark results for Super VL's vision-language tasks, real throughput for both Bolt models on 24GB and 4GB hardware, and how quickly the open weights are picked up by enterprises and developers. Those results will decide whether the efficiency pitch turns into shipped deployments.
Why it matters
NVIDIA is pushing open-model capability from the data center down to desktops and edge devices, which makes deployment cost and hardware footprint the next battleground for always-on agents.
Nearby Updates
All10/15, 23:02
Rivian plans to run third-party AI agents in its cars, with details promised in December
At TechCrunch Disrupt, Rivian founder and CEO RJ Scaringe said the company's future vehicles will be able to run AI agents built by other firms such as Anthropic, though he gave no timeline and the company pointed to its December "AI & Autonomy Day" event. Rivian has already rolled out an AI assistant and an in-house agentic framework that routes voice requests to the right part of its in-car software.
10/15, 22:29
Google's Universal Cart shopping tool launches in the US
Google has launched its Universal Cart shopping tool in the United States, available across Google Search and the Gemini app. The rollout plants a shopping cart inside both the search box and Google's AI assistant instead of a standalone shopping app.
10/15, 22:04
ServiceNow Rolls Out No-Code AI Agent Tools for Enterprise Employees
ServiceNow has unveiled a set of no-code AI agent tools, according to a GuruFocus report, aimed at letting employees build and use AI agents without writing code. It is another step in enterprise software pushing agent capabilities from engineering teams out to business users.
10/15, 21:00
NVIDIA Brings DLSS 4.5 Controls to GeForce NOW Cloud Gaming
NVIDIA has brought new DLSS 4.5 controls to GeForce NOW, giving cloud members more choice over how supported games look and perform. The company says the controls work even without the latest hardware because rendering happens on its servers, and the same update added Keen Games' Enshrouded 1.0 release to the service.