NVIDIA opens Nemotron 3.5 Super VL (120B-A12B multimodal agents) and Bolt 74B-A9B 24GB for local agents; Bolt E4B due Oct 23

Why it matters

【Key points】

On 2026-10-15 15:00Z NVIDIA released new open Nemotron 3.5 models: flagship multimodal agent model Nemotron 3.5 Super VL (~120B total / ~12B active) plus edge/local Nemotron 3.5 Bolt 74B-A9B 24GB; smaller Bolt E4B 4GB is slated for Oct 23. The corporate blog positions Super VL for long-running always-on agents (instruction following, tool use, multi-step workflows, visual document understanding) and claims up to ~5x throughput vs peers in class; Bolt uses native ~2-bit quantization to fit ~24GB and run agent workloads on RTX / DGX Spark locally. Weights are on Hugging Face (BF16/NVFP4 etc.), OpenRouter, ModelScope, and as NVIDIA NIM. Prior patrols watched “Bolt on HF”; this window delivers Super VL + Bolt in one official announcement.

【Verifiable】

Corporate blog (primary): https://blogs.nvidia.com/blog/nemotron-3-5-super-vl/ (pubDate 2026-10-15 15:00:24 +0000)

Developer tech blog: https://developer.nvidia.com/blog/nvidia-nemotron-3-5-super-vl-powers-specialized-agents-with-multimodal-reasoning/

Official X @NVIDIAAI: https://x.com/NVIDIAAI/status/2110749836450713896 (15:09Z)

HF Super VL BF16: https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-BF16 (~15:19Z)

HF Bolt 74B-A9B 24GB: https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Bolt-74B-A9B-24GB (~15:29Z)

ModelScope collection: https://modelscope.ai/collections/nv-community/Nemotron-35-Super-VL

【Our read】

1. Hard open multimodal agent release: corporate blog, developer blog, official X, and HF align — home_week.

2. Close the prior “Bolt watch”: Bolt 74B-A9B shipped with Super VL; E4B waits until Oct 23.

3. Marketing claims (5x, PinchBench #1, etc.) are vendor-stated; third-party replication still pending.

4. Customer customization list is adoption color, not separate product news.