Nemotron 3.5 Lightning

SWE-Bench Verified: 51.56% (BF16) / 52.80% (NVFP4), PinchBench: 85.37%, GPQA Diamond: 75.44%, MMLU Pro: 81.94%, Terminal-Bench 2.1: 24.58%, HLE: 11.72%, IFBench (loose): 71.88%, BrowseComp: 36.97%, AA Intelligence Index: 23.6, AA Coding Index: 26.8, AA Agentic Index: 13.8, AA Non-Hallucination Rate: 62.4% | NVIDIA's highest-efficiency model, purpose-built for the execution layer of always-on agents (released Aug 11, 2026). Distilled from Nemotron 3 Ultra. Designed for harnesses like OpenClaw and Hermes Agent. 4× output speed and 30% faster task completion vs similar-sized models (e.g. Qwen3.6-35B). Runs on a single DGX Spark or H100 GPU in NVFP4. 6 natural languages + 43 programming languages. OpenMDW 1.1 license, fully open (weights, data, recipes). **Best for:** high-volume agentic execution, repetitive tool calls/validation/formatting, sub-agent workhorse in multi-model pipelines, coding agent tasks at extreme throughput and ultra-low cost (~$0.05/$0.20 per MTok). **Not for:** autonomous planning, hard reasoning, broad terminal autonomy, or any task requiring frontier intelligence — this is an execution model, not a thinking model. Terminal-Bench 2.1 at 24.58% and HLE at 11.72% confirm the scope.**