Artificial Analysis lists NVIDIA's Nemotron 3 Nano
Artificial Analysis has added NVIDIA's Nemotron 3 Nano 30B A3B to its tracker: an open-weights mixture-of-experts model with 3.6B active parameters, a 1M-token context window, and a very low listed price.
Artificial Analysis has published a model page for NVIDIA's Nemotron 3 Nano 30B A3B. It covers two variants, one reasoning and one non-reasoning.
Both variants share the same architecture specs. The model has 31.6B total parameters, of which 3.6B are active per token at inference. It has a 1M-token context window and handles text input and output only. The weights are on Hugging Face under the Nvidia Nemotron Open Model License. Because of its size, Artificial Analysis puts the model in its "small" open-weights class (4B–40B parameters) and compares it only against other open-weights models in that range.
The page lists both variants at $0.065. Artificial Analysis generally reports prices per million tokens, blended at a 3:1 input-to-output ratio. The non-reasoning variant has the lowest time to first answer token of the two, at 0.87 seconds. The comparison table also shows figures of 188 for the reasoning variant and 167 for the non-reasoning variant, but the extracted page doesn't label what they measure.
The page names the evaluations behind its capability indexes, including AA-Omniscience v1.1, AA-Briefcase v1.1, HLE-Diamond, AutomationBench-AA, Terminal-Bench 4.0 and FrontierCode-AA 2.0. The available material doesn't include the model's scores on any of them, its Intelligence Index rating, or cost per task. It also gives no release date or NVIDIA commentary.