NVIDIA AI Updates: August 5, 2026
1. NVIDIA Releases Alpamayo 2 Super, an Open Reasoning Model for Autonomous Vehicles
NVIDIA. NVIDIA made Alpamayo 2 Super available for commercial use, an open reasoning model built for robotaxis and autonomous vehicles. The 34-billion-parameter model pairs a 32B Cosmos 3 Super Reasoner with a 2B diffusion-based Action Expert to generate driving trajectories alongside chain-of-causation reasoning traces, predict meta-actions like yielding or lane changes, and answer natural-language questions about driving scenes with 2D visual grounding. NVIDIA reports it ranks first among nearly 40 models on autonomous driving benchmarks while remaining deployable for on-vehicle inference. Source
2. NVIDIA Introduces Cosmos 3 to Power World Action Models for Robot Manipulation
NVIDIA. NVIDIA released Cosmos 3, an open foundation model built to power World Action Models that predict how the physical world evolves rather than just describing scenes, unlike Vision-Language-Action models. The Mixture-of-Transformers model was trained on 767 million images, 348 million videos, and 8 million action samples across multiple domains. NVIDIA says the dynamics-aware approach lets robotics policies generalize to new environments and embodiments with fewer demonstrations, and released trained policy models such as Cosmos3-Nano-Policy-DROID alongside it. Source
3. NVIDIA Open-Sources cuFile APIs and Launches Storage-Next Initiative
NVIDIA. NVIDIA open-sourced its cuFile APIs, which let GPUs access storage directly with microsecond latency, and launched Storage-Next, an initiative with more than 40 storage vendors to establish GPU-driven storage standards. The announcement, made at the Future of Memory and Storage conference, also cited Vera CPU benchmarks showing 3.21 times higher throughput than an x86 CPU on compression and encryption tasks. Google, Intel, and Meta join NVIDIA as inaugural maintainers of the open cuFile project. Source
4. Open Secure AI Alliance Proposes SAFE Guidelines for Sharing AI Security Incidents
NVIDIA. The Linux Foundation, on behalf of the 120-plus-member Open Secure AI Alliance, proposed Shared AI Findings Exchange guidelines to confidentially collect and analyze AI security incidents and near misses and publish evidence-based recommendations. NVIDIA is leading technical contributions across the security stack, including agent harnesses, runtime controls, and specialized models. The framework targets collective, transparent defense against agentic AI threats. Source
5. NVIDIA Joins NSF State and Regional AI Hubs Program
NVIDIA. NVIDIA joined the National Science Foundation’s State and Regional AI Infrastructure Hubs program, a new initiative connecting government, academia, and industry to expand access to advanced computing, data, software, and expertise for AI research and education across the United States. NVIDIA is contributing technology, implementation expertise, and workforce development resources, building on its earlier partnership with the University of Florida. The program aims to build regional talent pipelines tied to priorities such as healthcare, energy, and quantum computing. Source