Newmotron 3.5 Lightning: Specially for Agent

In Weida, an open source model, Nemotron 3.5 Lightning, was launched for the use of tools specifically for Agent, which is running for a long time. The model ' s total parameter is 30 billion, with only 3 billion tokens activated per token, and is primarily responsible for high-frequency tasks such as tool calls, results checks and subAgent movements. The idea is to get Agent's division of labour more thoroughly, to hand over complex planning to large models such as Nemotron 3 Ultra, and to give a lot of repetitive execution to Lightning. Officially, it can be up to four times as fast as the equivalent model. According to artificial intelligence analysis, Nemotron 3.5 Lightning performed well in an independent assessment at a speed of 416 token/s, ranking first in the same size open weight model. Despite a 24-point smart index, the performance in Agent ' s capacity development was significant, with ELO reaching 824 and exceeding gpt-oss-120b and Nemotron 3 Super。