The Open-Source AI Stack
RSS
All models

Models · Compare

Llama-3.3-Nemotron Super 49B v1 vs Gemini 2.5 Pro

Rows highlighted in warm gray are where the models differ. Numbers carry their as-of date and primary source.

Specs

Field A: Llama-3.3-Nemotron Super 49B v1 B: Gemini 2.5 Pro
Released 2025-03-182025-03-25
Developer NVIDIAGoogle DeepMind
Openness Open weightsProprietary
License NVIDIA Open Model LicenseProprietary
OSI-approved nono
Data released yesno
Training code nono
Architecture denseunknown
Total params 49B
Active params
Experts
Context window 131K1.0M
Attention skip-attentionunknown
Position enc. ropeunknown
Pretraining tokens 40B
Post-training sft, grporlhf
Training hardware H100
$/M input $0.00$1.25
$/M output $0.00$10.00
Output tok/sec 0127.3

Benchmarks

Missing scores render as not reported; never inferred. Bold highlights the leader per benchmark.

General reasoning

MMLU-Pro 69.8 2026-05-21 86.2 2026-05-21
GPQA-Diamond 66.7 2025-03-18

Code

LiveCodeBench 28.0 2026-05-21 80.1 2026-05-21

Math

MATH 96.6 2025-03-18 96.7 2026-05-21
AIME 2024 19.3 2026-05-21
AIME 2025 58.4 2025-03-18 87.7 2026-05-21

Held-out / arena

IFEval 89.2 2025-03-18

Context · A

NVIDIA's NAS-distilled Llama 3.3 70B aimed at single-data-center-GPU throughput. Uses skip-attention and variable-FFN blocks selected per layer for the quality-versus-FLOPs tradeoff. Released March 18 2025 with the Llama-Nemotron-Post-Training-Dataset-v1 (30M samples) public.

Context · B

The first Gemini release to clearly lead on LMArena Elo and on hard reasoning benchmarks. Native 1M-token context with reported 2M expansion in pipeline.

Llama-3.3-Nemotron Super 49B v1 detail → · Gemini 2.5 Pro detail →