NVIDIA

Open reasoning and agentic models from the company that builds the hardware

NVIDIA designs the GPUs most of this industry trains and serves on, and it publishes open models of its own. The Nemotron family covers reasoning, tool use and long-context work at sizes from a few billion parameters upward, with open weights and permissive terms, and the smaller members are tuned to fit a single card. Through Runware you deploy them on your own workers and pay for the hardware they hold.