Nemotron 3.5 Super VL
AvailableNVIDIA Nemotron 3.5 Super VL (NVIDIA-Nemotron-3.5-Super-VL-120B-A12B) is an open vision-language model released 2026-10-15 for always-on and multimodal agent workloads. Hybrid Mamba-2 / Transformer MoE (512 experts per MoE layer), 120B total / 12B active, C-RADIO vision encoder; text, image and video (up to 2 min) in, text out; up to 1M-token context. Weights released (incl. NVFP4) under OpenMDW-1.1, alongside open datasets, training recipes and NeMo Gym RL environments; runs on a single DGX Spark / H200 / GB200. Model card (NVFP4): GPQA Diamond 89.9, SWE-bench Verified 71.07, Terminal-Bench 2.1 62.78, OSWorld-Verified 52.3. Artificial Analysis Intelligence Index 26.5.
Specifications
- License
- Open weights · OpenMDW-1.1
- Weights
- Downloadable
- Architecture
- hybrid
- Parameters
- 120B · 12B active
- Context window
- 1M tokens
- Max output
- —
- Knowledge cutoff
- —
- Price (in / out, $/M)
- —
- Modalities
- TextVisionVideoCode
Benchmarks
No benchmark scores recorded yet. Spotted some? Submit a correction.
Vendor-reported figures are claims until independently verified. See methodology.