Nvidia announced its Nemotron 3.5 Lightning, a 30‑billion‑parameter Mixture‑of‑Experts (MoE) model, alongside the open‑source NeMo Switchyard router. Posts on X relaying the launch named Cognition among the ecosystem partners, saying it cut Devin Desktop costs with the router.
According to a post by @GenAISpotlight on X, Nemotron 3.5 Lightning is built for high‑volume agentic workflows and delivers four times faster output generation, completing agentic tasks about 30 % faster than comparable models. The post also notes that the model runs on RTX‑based PCs, DGX Spark systems and data‑center hardware.
The same user described NeMo Switchyard as a dynamic router that directs each agent step to the most suitable model based on cost, latency and quality, preserving frontier accuracy while cutting task execution costs by roughly 67 % relative to running Opus 4.8 alone.
A separate X post by @Synapse_Brief said Cognition wired the router into Devin Desktop and achieved “near frontier results at 28 % lower mean cost.” The user added that the integration also yielded faster task execution, though specific speed figures were not provided.
Sources
Fan coverage from Devin Central — not an official Cognition announcement. Devin is a trademark of Cognition.