DATASHEET
NVIDIA/RELEASED Oct 15, 2026
Nemotron 3.5 Super VL
120B OPEN HYBRIDNVIDIA's new open agentic vision-language model: 120B/12B-active Mamba-2-MoE hybrid with a C-RADIO vision encoder for image, document, and video understanding, 1M context, native MTP speculative decoding, and full open weights (BF16 + NVFP4) under OpenMDW-1.1; Terminal-Bench 2.1 60.67%, SWE-bench Verified 71.1%, GPQA-Diamond 89.52%, at $0.15/$0.50 per 1M tokens.
Context
1,048,576 tokens
Architecture
120B Total / 12B Active MoE (Mamba-2 hybrid, 512 experts, C-RADIO vision)
Official API (1M)
$0.15 in / $0.5 out (Open)+ Weights Free to Self-Host
License
OpenMDW-1.1
Published benchmarks · lab-reported
Vendor-published figures, not independently reproduced. Sources in Verification below.
SWE-BENCH
71.1% (Verified)
TERMINAL-BENCH
60.67% (Terminal-Bench 2.1)
GPQA DIAMOND
89.52% (GPQA-Diamond)
Quickstart
curl -s https://modelregistry.tirup.in/api/cli?model=nemotron-3-5-super-vlSources
- Nemotron 3.5 Super VL model card · NVIDIA
- Nemotron 3.5 Super and Lightning technical report · NVIDIA
- Nemotron 3.5 Super VL listing and pricing · OpenRouter