SHELFCRITTER AI Digest

MODEL

Nemotron Bolt

modeltopic-notenvidianemotronopen-weightson-device

Overview

Nemotron Bolt is a pair of small open-weight models in NVIDIA‘s Nemotron line, launched on 2026-10-19 and framed as the on-device / consumer-GPU tier of the 3.5 generation alongside the larger Nemotron 3.5 Super VL enterprise-agent model. Two variants: a 74B-A9B configuration quantized to 2-bit so it fits in 24GB VRAM, and an E4B variant at 4GB for conversational agents on RTX hardware and DGX Spark.

Timeline

  • 2026-10-27-AI-Digest — Both Bolt variants shipped alongside Nemotron 3.5 Super VL. The honest read the corpus carries: credible frontier-class open weights are now landing on local hardware with an explicit agent-stack framing, extending the open-weight-closing-ground arc on on-device agent tasks — though the coding-frontier gap persists.

Key Developments

  1. Two on-device variants targeting RTX / consumer-GPU footprints, launched October 19, 2026: The 74B-A9B at 2-bit fits a 24GB card (prosumer RTX / RTX Ada range); the E4B at 4GB runs on mainstream RTX hardware and DGX Spark. The variant-split is NVIDIA’s way of covering both the “powerful local agent” and “chatty assistant on a laptop” footprints with one release cadence.

  2. Agent-stack framing, not just weights: Like other recent Nemotron releases, Bolt ships with explicit positioning for agent workloads rather than as a bare weights drop — the on-device framing is deliberate, aimed at the practitioner who wants a frontier-adjacent open-weight model running on their own hardware.

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