header-langage
简体中文
繁體中文
English
Tiếng Việt
한국어
日本語
ภาษาไทย
Türkçe
Scan to Download the APP
Partnered With:

NVIDIA Nemotron 3.5 Lightning Benchmark: Strong in Speed, Falls Short in Intelligence

According to MotionWatch Beating Monitor, Artificial Analysis has completed the independent evaluation of Nemotron 3.5 Lightning. The model has a total of 31.6 billion parameters, with 3.6 billion active parameters. It received an intelligence quotient of 24 points, on par with the OpenAI gpt-oss-120b model, which has approximately 4 times the parameters, and scored 9 points higher than the previous generation Nemotron 3 Nano.

One of Lightning's most prominent features is its speed. Prior to the Artificial Analysis release, it achieved close to 670 Token/s on DeepInfra testing; it currently stands at 416 Token/s on the official leaderboard, ranking first among 134 comparative models of the same size in terms of open-weight models.

There is also a significant improvement in the Agent's capabilities. On GDPval-AA v2, its Elo reached 824, surpassing both gpt-oss-120b and Nemotron 3 Super; on Terminal-Bench v2.1, the performance increased from 7% in the previous generation to 24%, more than a threefold improvement.

However, it is not the most intelligent model of its size. In terms of the Artificial Analysis intelligence quotient, Qwen3.6 35B A3B scored 32 points, and Muse Glimmer High scored 35 points, while Lightning scored only 24 points. NVIDIA did not pursue the highest intelligence score for the model of the same size; instead, they focused on speed and Agent execution.

举报 Correction/Report
RelatedRecommended
The most anticipated annual event in the AI field, NVIDIA's GTC 2026 Conference, kicked off today in San Jose, California, USA.
Some estimates suggest that losing its founder could shrink Nvidia's valuation by 20% to 30%.
Is Nvidia finished? Export controls, RASA, open-source impact, supply chain impact, extramarital affairs, the US still relies on China.
Correction/Report
Submit
Add Library
Visible to myself only
Public
Save
Choose Library
Add Library
Cancel
Finish