Nemotron 3.5 Lightning: benchmark results
NVIDIA's Nemotron 3.5 Lightning, a 30B-A3B hybrid Mamba-2/MoE open model released in full BF16 precision as the reference checkpoint for further fine-tuning, distillation or quantization. Provider: NVIDIA. Released 2026-08-11. Access: Open.
Aggregate Rating 1592 ± 12, rank #471 of 1656 rated models, from 234 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Liquid LFM2.5 24B Card - AA-Omniscience | -3.58 | Omniscience index (−100 to100) | 100 |
| Voice Standard v0.1.3 (gpt-5.6-sol; 120s) - Information Extraction | 100 | Published information_extraction task_quality (1-100), canon | 100 |
| ToolSandbox | 75.13 | Avg Score (imported) | 93.8 |
| AA-Omniscience Hallucination Rate | 37.57 | Hallucination Rate (%) | 88.1 |
| Medmarks - Med-HALT Reasoning NOTA | 69.19 | Score (%) | 87.5 |
| Medmarks - SCTpublic | 71.86 | Score (%) | 86.4 |
| Medmarks - Med-HALT Reasoning FCT | 86.5 | Score (%) | 78.4 |
| Medmarks - LongHealth Task 2 | 88.92 | Score (%) | 77.3 |
| SpeechMap Compliance | 75 | % Requests Completed | 71 |
| Medmarks - MetaMedQA | 74.92 | Score (%) | 69.3 |
| MLCR-AA - Conciseness | 97.22 | Conciseness (%) | 67.9 |
| Voice Standard v0.1.3 (gpt-5.6-sol; 30s) - Summarization | 98 | Published summarization task_quality (1-100), canonical v0.1 | 67.6 |
Interactive version: theaggregate.ai/model?slug=nemotron-3-5-lightning · How It Works · Data refreshed daily, snapshot 2026-10-21.