Llama 3.3 Nemotron Super 49B v1 (Reasoning) — benchmark results

Llama 3.3 Nemotron Super 49B v1 evaluated with reasoning enabled. Provider: NVIDIA. Released 2025-03-18. Access: Open.

Unified ELO 1493 ± 44, rank #826 of 1776 rated models, from 37 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
AA MATH-50095.87Accuracy (%)81.7
AA MMLU-Pro78.46Accuracy (%)60.5
AA AIME 202554.67Accuracy (%)51.1
AA Omniscience - Software Engineering (SWE) - Java17Accuracy (%)51
AA Omniscience - Software Engineering (SWE) - JavaScript29.09Accuracy (%)48.9
AA Omniscience - Health18.6Accuracy (%)47.9
AA Omniscience - Software Engineering (SWE) - HTML30Accuracy (%)47.8
AA Humanity's Last Exam6.52Accuracy (%)47.7
AA GPQA Diamond64.34Accuracy (%)43.4
AA Omniscience - Science, Engineering & Mathematics24.1Accuracy (%)41.5
Artificial Analysis Intelligence Index12.25Intelligence Index41.3
AA Omniscience - Business14.7Accuracy (%)39.5

Interactive version: theaggregate.ai/model?slug=llama-3-3-nemotron-super-49b-v1-reasoning · How the rankings work · Data refreshed daily, snapshot 2026-07-22.