NVIDIA Nemotron 49B v1 Card - Arena-Hard: leaderboard
Arena-Hard: difficult real-user prompts whose answers an LLM judge compares with a baseline model's answers; the card prints only a reasoning-off score. From NVIDIA's own Llama-3.3-Nemotron-Super-49B-v1 model card, which states a 32k sequence length with benchmarks run up to 16 times and averaged, and recommends temperature 0.6 with top-p 0.95 for reasoning on and greedy decoding for reasoning off. NVIDIA's Llama-Nemotron report prints the same value labelled reasoning on (NVIDIA Nemotron 49B Report board), so the two sources' mode labels disagree.
Metric: Score. Source: huggingface.co. 1 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Llama 3.3 Nemotron Super 49B (v1) (Non-reasoning) | 88.3 | #1336 (Llama 3.3 Nemotron Super 49B (v1)) |
Interactive version: theaggregate.ai/benchmark?slug=nvidia-nemotron-49b-v1-card-arena-hard · How It Works · Data refreshed daily, snapshot 2026-11-12.