Nvidia is training a trillion-parameter model it plans to give away — and it is already behind on size
The Information reported on 11 August, citing employees working on the project, that the largest model in Nvidia's next open family, Nemotron 4, will carry at least a trillion parameters — roughly twice the 550 billion of Nemotron 3 Ultra. Training is not finished and there is no announced release date, though staff said it could be ready as early as late autumn. Nvidia is one of the few large American companies still publishing open weights, and the stated aim is to give enterprises a Western model they can download and run on their own hardware rather than renting through an API. The number is not a record: Alibaba published Qwen3.8-Max at 2.4 trillion parameters this week, and Moonshot's Kimi K3 arrived at 2.8 trillion in July. Nvidia has not commented publicly on the plan.
Related stories
DeepSeek's cheap workhorse can now see — and on agent tasks that need eyes it says it is close to Anthropic's best2026-08-21Show the robot once — three to twelve seconds — and it gets the job right 59 times out of 100 with no training at all
2026-08-21The model invents its own tasks, builds the rig to test them, then trains on the results — and DeepReinforce gave the weights away
2026-08-20A free model that fits on one graphics card scores 52 — then spends 22,276 thinking tokens drawing a picture
2026-08-18Zhipu did not build a new model. It kept training the old one — and says coding got 50 percent better.
2026-08-17