ON JANUARY 6TH, DURING THE CES 2026 KEYNOTE SPEECHNvidiaChief Executive OfficerJen-Hsun HuangA keynote address to announce a massive expansion of itsOpen SourceModel librariesRelease of new models covering the four main areas of language, robotics, autopilot and medicineDatasetFURTHER ACCELERATION OF INDUSTRY-WIDE AI INNOVATION。
The contribution of YVD to the Open Source Training Framework and the world's largest open multi-modular data set, including 10 trillion language training tokens, 500,000 robot tracks, 455,000 protein structures and 100TB vehicle sensor data, marked YVD's efforts to build an open ecosystem encompassing language processing, robotic technology, scientific research and self-driving. The following video is attached to the IT home:
Many technology giants, including Bosch, Salesforce, Uber and Palantir, now use these open-source technologies to build their next generation AI systems。
This release includes the Nemotron series for smart AI, the Cosmos platform for physical AI, the Alpamayo series developed for autopilot and the Clara model in biomedical fields。
Nemotron Enabling Intelligence AI, voice recognition 10 times higher
In the area of smart AI, NVIDIA has introduced a new Nemotron series model, covering the three main sectors of voice, retrieval enhancement generation (RAG) and security。
Of these, the Nemotron Speech model, which has performed well in real-time subtitles and voice applications, has been used by Boshi to optimize the interactive experience of vehicle-borne voice, as benchmark tests show that it works 10 times faster than similar models。
At the same time, the Nemotron Safety model has significantly increased the confidence of enterprise-level AI applications by enhancing content security testing and sensitive data recognition, and has been adopted by security companies such as CrowdStrike and Fortinet。
Cosmos joined forces with Isaac to give robotic "physical world" reasoning power
For physics AI (Physical AI), NVIDIA launched the Cosmos World Model Platform, which aims to give robotics the same human-like reasoning and world-generated power。
The core model Cosmos Reason 2 has significantly improved robotic perception and interactive precision of the physical environment, while CosmosTransfer 2.5 produces large-scale synthetic video to train AI。
Based on this platform, NVIDIA also launched the Isaac GR00T N1.6 model, designed for human robots, with full body control and environmental reasoning。
Companies such as Franka Robotics are now using these tools to validate robotic behaviour in the virtual environment and then deploy it to the real world。
Launch of the Alpamayo series, reshaping reasoning autopilot development
For the first time, NVIDIA launched the Alpamayo series of open source resources to tackle the problem of autopilot. Among them, Alpamayo 1 is the first large-scale open-source reasoning VLA (visual language action) model for automatic driving, which not only allows a vehicle to understand the environment but also explains its driving decisions. In conjunction with the Open Source Simulation Framework, AlpaSim, developers can conduct closed-ring training to cope with peripheral scenarios。
IN ADDITION, NVIDIA HAS AN OPEN SOURCE OF A PHYSICAL AI DATA SET CONTAINING OVER 1,700 HOURS OF DRIVING DATA, COVERING AN EXTREMELY WIDE GEOGRAPHICAL ENVIRONMENT AND COMPLEX ROAD CONDITIONS, AND PROVIDING CRITICAL DATA SUPPORT FOR HIGH-LEVEL AUTO-DRIVING RESEARCH AND DEVELOPMENT。
Clara Model goes deep into the microworld and accelerates the development of new drugs
In the area of health care, NVIDIA has introduced a new Clara AI model to shorten the drug development cycle. The La-Proteina model supports the design of proteins of atomic-grade precision to help scientists cope with difficult diseases; ReaSyn v2 integrates the blueprint of manufacturing into the development process to ensure that the drugs designed are synthetic。
IN ADDITION, KERMT MODELS CAN IMPROVE SAFETY BY DEVELOPING EARLY PREDICTIONS OF THE INTERACTION OF DRUGS WITH HUMANS. TOGETHER WITH THE NEWLY RELEASED 455,000 SYNTHETIC PROTEIN STRUCTURE DATA SETS, THESE TOOLS WILL EFFECTIVELY REDUCE THE THRESHOLD AND COST OF MEDICAL INNOVATION。