NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-BF16 model card

Primary card for the Nemotron 3.5 Super VL record. HF API createdAt 2026-10-12; the card's own Release Date line says 2026-10-15. Card states 120B total / 12B active, Mamba-2 plus MoE plus attention hybrid with a C-RADIO vision encoder, context up to 1M tokens, OpenMDW-1.1. Benchmark table on the card (GPQA Diamond 89.52, Terminal-Bench 2.1 60.67, SWE-bench Verified 71.10, SWE-bench Pro 51.78, with the same columns for Qwen3.5-122B-A10B, Qwen3.8-Flash-Next, DeepSeek-V4-Flash-0731, Inkling-Small and GLM-5.3-Flash) matches Table 10 of the technical report and is a vendor claim, not an independent re-run. Deployment section names vLLM and TensorRT-LLM cookbooks and an NVFP4 sibling checkpoint. The routed-expert count is not on the card; that number comes from config.json.