SHELFCRITTER AI Digest

MODEL

Nemotron 3.5 Super VL

modeltopic-notenvidianemotronopen-weightsmultimodal

Overview

Nemotron 3.5 Super VL is an open-weight multimodal model in NVIDIA‘s Nemotron line, launched on 2026-10-19 and framed as the always-on enterprise-agent entry in the 3.5 generation (alongside the sparser Nemotron 3.5 Lightning on-prem/edge variant). Spec: 120B total parameters / 12B active, open weights. NVIDIA’s launch claim is that Super VL matches its largest internal reasoning model at roughly 1/5 the parameters and is up to 5× faster than rivals on Artificial Analysis’s leaderboard.

Timeline

  • 2026-10-27-AI-Digest — Nemotron 3.5 Super VL (120B / 12B-active) shipped alongside the smaller Nemotron Bolt variants as a direct play against proprietary agent models. NVIDIA named Skydance, EY, Ramp, CrowdStrike, Perplexity, and Salesforce as early users — treated by the corpus as marketing framing (“companies including…”), not independent partner confirmation. The honest read: credible frontier-class open weights are now on local hardware with an explicit agent-stack framing; the open-weight-closing-ground arc extends on multimodal and on-device agent tasks but the coding-frontier gap persists (Aider polyglot top-5 has no open-weight entries today).

Key Developments

  1. Open-weight multimodal agent model, 120B/12B-active, launched October 19, 2026: NVIDIA’s framing positions Super VL as the “always-on enterprise agent” tier of the 3.5 family, differentiated from the sparser Lightning variant pitched at the RTX-through-DGX serving fabric. The release continues NVIDIA’s pattern of pairing open-weight model releases with an explicit agent/inference-stack framing rather than shipping bare weights.

  2. Six-logo launch slate is Nvidia-sourced, not independently confirmed: The Skydance / EY / Ramp / CrowdStrike / Perplexity / Salesforce list carries Nvidia’s own characterization (“companies including…”), which is a mix of evaluation, pilot, and production — practitioners should read it as a launch-slide slate until any of the six publishes its own characterization.

See also: Nemotron, Nemotron Bolt, Nemotron 3.5 Lightning, NVIDIA, MOC - Open Source Models, MOC - Agentic Coding.