OpenAI has unveiled GPT-5.2, the company’s most advanced model series yet, designed for professional knowledge work. Built and deployed on NVIDIA’s full-stack AI infrastructure—including Hopper and the latest GB200 NVL72 systems—GPT-5.2 exemplifies the growing reliance of leading AI developers on purpose-built hardware and software platforms to train and scale frontier models. The model’s development underscores the critical role of infrastructure in enabling the next generation of artificial intelligence. At the heart of modern AI advancement are three scaling laws: pretraining, post-training, and test-time scaling. While reasoning models that use compute during inference to solve complex problems are now widespread, pretraining and post-training remain foundational. These stages are essential for building intelligence, enabling models to understand context, logic, and nuance. Training such models from scratch demands immense computational power—tens or even hundreds of thousands of GPUs operating in concert across scale-up, scale-out, and scale-across architectures. Success requires not just powerful accelerators but also high-speed networking and a fully optimized software stack, all delivered through a unified, high-performance platform. NVIDIA’s GB200 NVL72 systems have demonstrated a 3x speedup in training performance over the previous Hopper architecture on the largest models tested in the latest MLPerf Training benchmarks, with nearly 2x better performance per dollar. The upcoming GB300 NVL72 offers more than a 4x improvement over Hopper, drastically reducing development cycles and accelerating time-to-deployment for new models. NVIDIA’s platform supports AI across all modalities. Beyond text, it powers breakthroughs in speech, image, video, biology, and robotics. Models like Evo 2 decode genetic sequences, OpenFold3 predicts 3D protein structures, and Boltz-2 simulates drug interactions to accelerate pharmaceutical discovery. In healthcare, NVIDIA Clara generates synthetic medical images for diagnosis and screening without compromising patient privacy. Creative industries are also leveraging the platform. Runway recently launched Gen-4.5, the top-rated video generation model globally, trained entirely on NVIDIA GPUs. It’s now optimized for the NVIDIA Blackwell architecture. Runway also introduced GWM-1, a general world model capable of simulating reality in real time. Designed to be interactive, controllable, and general-purpose, GWM-1 has applications in gaming, education, science, entertainment, and robotics. NVIDIA’s dominance in AI infrastructure is reflected in industry benchmarks. In the latest MLPerf Training 5.1 results, NVIDIA was the only platform to submit across all seven benchmarks, showcasing unmatched versatility and performance. This efficiency enables AI labs—such as Black Forest Labs, Cohere, Mistral, OpenAI, Reflection, and Thinking Machines Lab—to train models faster and more cost-effectively. NVIDIA Blackwell is now widely available through major cloud providers including Amazon Web Services, Google Cloud, Microsoft Azure, Oracle Cloud, and others, as well as neo-clouds and server manufacturers. The NVIDIA Blackwell Ultra, with enhanced compute, memory, and architecture, is rolling out to support even more demanding workloads. From frontier AI to everyday applications, the future of artificial intelligence is being built on NVIDIA’s platform. As models grow more capable and complex, the need for scalable, high-performance infrastructure becomes even more critical. GPT-5.2’s launch, powered by NVIDIA’s technology, marks a significant milestone in the evolution of AI—demonstrating that the most advanced models are not just the result of better algorithms, but of a complete, integrated system built for scale, speed, and innovation.