NVIDIA ships Nemotron 3.5 Super & Lightning: 120B MoE VLM scores 51.78% on SWE-bench Pro
gordic_aleksa · x · 2026-10-15
NVIDIA's Nemotron pretraining team announced Nemotron 3.5 Super and Lightning:
- Super: a 120B total / 12B active vision-language model scoring 51.78% on SWE-bench Pro, beating Nemotron 3 Ultra with 4.6× fewer active params
- Lightning: a 30B total / 3B active text-only LLM
- Architecture: MoE Hybrid Mamba-Transformer built for efficient inference, 1M token context window, native speculative decoding via MTP
- Training/deployment: trained on GB200/GB300, validated across all imaginable SKUs including consumer RTX and DGX Spark
- Benchmark charts show Super leading Qwen3.5, DeepSeek-V4-Flash, and GLM-5.3 Flash on SWE-bench Multilingual, Terminal-Bench, BrowseComp and more, with max output throughput of 4,331.68 TPS/GPU on GB200 — 2.34× Qwen3.5
More from Models
- OpenAI ships GPT-Live-1.1: new voices in six languages, fewer confirmations — OpenAI · 2026-10-16
- Muse Spark model appears on LMArena, testable in Battle and Agent modes — arena · 2026-10-16
- Running a multi-model UI design arena: Paseo orchestrates four design drafts per model — vista8 · 2026-10-16
- Ilya Sutskever questions Meta license clause taxing MaaS providers above $20M revenue — ilyasu · 2026-10-16
- Muse-Spark open-weights drop: 816B-A42B MoE model gets 359GB MLX 3-bit build — ns123abc · 2026-10-16
- Midas flags inconsistent Claude limits: 33% weekly and 45% window usage despite $100 credits, yet out of API credits — TheMoonMidas · 2026-10-16