Nemotron Super 49B
Single-GPU 49B reasoning model pruned from Llama 3.3 70B
About
Nemotron Super is the 49B member of NVIDIA's Llama Nemotron family, created by pruning Llama 3.3 70B with neural architecture search and then distilling it down to fit on a single data-center GPU. NVIDIA positions it as the throughput pick: best accuracy you can serve without multi-GPU infrastructure, with a reasoning mode for hard problems and a fast mode for chat. Reported numbers include 96.6% on MATH500 with reasoning on and strong instruction-following scores. Like the rest of the family it uses NVIDIA's Open Model License rather than an OSI license. It is still a 49B model, so "single GPU" means a data-center card, not a laptop.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
NVIDIA-reported: MATH500 96.6 (reasoning on) vs 74.0 (off); MT-Bench 9.17 reasoning-off
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 8/10
Scored 8/10 (verified tier, metrics-v1). Evidence: Part of NVIDIA Llama Nemotron family (March 18 2025): 49B pruned from Llama 3.3 70B via NAS, single-GPU serving.; Benchmarks NVIDIA-reported: MATH500 96.6 reasoning-on, 74.0 off; MT-Bench 9.17.; License caveat: NVIDIA Open Model License, not OSI-approved.; Limitation: 'single GPU' means data-center class hardware..