Nemotron-Flash-3B: benchmark results

Provider: NVIDIA. Access: Open.

Aggregate Rating 1394 ± 37, rank #6435 of 10552 rated models, from 26 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
Lara Mini Base Report (base-model protocol) - MATH (4-shot)67.7Math-Verify equivalence, Minerva prompts, 5,000 problems ove98.1
Lara Mini Base Report (base-model protocol) - MBPP+ (3-shot)58.7pass@1; must pass MBPP's own and the EvalPlus tests (%)88
Lara Mini Base Report (base-model protocol) - XNLI (English) (0-shot)55.2Accuracy (%), English benchmark87
Lara Mini Base Report (base-model protocol) - PIQA (0-shot)80Accuracy normalized by answer length (%), English benchmark86.1
Lara Mini Base Report (base-model protocol) - MMLU-Pro (5 (CoT)-shot)39.3Exact match (%), English benchmark85.2
Lara Mini Base Report (base-model protocol) - MBPP (3-shot)52.2pass@1 (execution) (%)79.6
Lara Mini Base Report (base-model protocol) - MGSM (8 (CoT)-shot)46.1Exact match on each language's native CoT answer phrase, ave75.9
Lara Mini Base Report (base-model protocol) - GSM8K (5-shot)65.4Exact match (strict extraction: first "#### N", else first \68.5
Lara Mini Base Report (base-model protocol) - ARC-Challenge (25-shot)57.3Accuracy normalized by answer length (%), English benchmark63
Lara Mini Base Report (base-model protocol) - Belebele (English) (0-shot)70.8Accuracy (%), English benchmark63
Lara Mini Base Report (base-model protocol) - PAWS-X, All Languages (0-shot)56Accuracy (%) averaged over all languages/variants of the ben63
Lara Mini Base Report (base-model protocol) - PAWS (English) (0-shot)65.6Accuracy (%), English benchmark62

Interactive version: theaggregate.ai/model?slug=nemotron-flash-3b · How It Works · Data refreshed daily, snapshot 2026-11-16.