NVIDIA Nemotron-3.5-Super-VL lands: 51.79% SWE-Bench Pro, beats Ultra-3.0 on most benchmarks
kuchaev · x · 2026-10-15
NVIDIA released Nemotron-3.5-Super-VL (120B total / 12B active, now with a vision encoder):
- Benchmarks: 51.79% on SWE-Bench Pro; Artificial Analysis index of 26 vs 23 for Ultra-3.0, with the official claim it leads Ultra-3.0 on most benchmarks and adds visual reasoning
- Post-training: a greatly simplified pipeline vs Ultra-3.0, with fewer teacher models and a smaller SFT set
- Architecture: same base as Super-3.0; tryable now via NIM API with thinking mode and image input
The release ships alongside fully open weights, datasets, training recipes and a technical report.
Related event: NVIDIA Open-Sources Multimodal Nemotron 3.5 Super VL(22 posts)→
More from Models
- Datadog's Scout 1.0: post-trained GLM-5.3 Flash matches frontier models for AI SRE at lower cost — Tim_Dettmers · 2026-10-16
- Nemotron 3.5 Super finishes 4x more enterprise agent jobs using 123x fewer tokens per task — ryanshrout · 2026-10-16
- NVIDIA ships two Nemotron-3.5 open models in one day: Super for H100, Bolt for RTX — kuchaev · 2026-10-16
- Fine-tuned Nemotron 3.5 matches Opus 4.7 on Harvey legal agent bench at 1/11 the price — simonguozirui · 2026-10-16
- ActICL trains MLLM agents to build their own ICL contexts, +11.6 points over retrieval — EliasEskin · 2026-10-16
- Google publishes its September 2026 AI updates roundup — OnlineInference · 2026-10-16