Unveiling what it describes as the most capable series yet for professional information work, Openai launched GPT-5.2 today. The model was trained and deployed on the Nvidia infrastructure, including the Nvidia hopper and GB200 NVL72 systems.
The latest example of how AI leaders are leading the way and deploying at scale is NVIDIA's complete AI infrastructure.
To be like: the bed of wisdom
AI models are getting more efficient thanks to three rules of measurement: simulation, post-training and time-testing.
Deliberative models, which use computational time to tackle complex questions, using multiple networks working together, are now everywhere.
But temper tantrums and postpartum training are always a bed of genius. They are the basis for making models that think intelligently and are very useful.
And getting there takes shape. Frontier Frontier models from scratch is no small task.
It takes tens of thousands, even hundreds of thousands, of well-coordinated GPUs.
That level of measure calls for a contrast of many dimensions. It requires world-class accelerators, advanced networks at all levels, scaling and increasingly reaching political structures, and well-developed software. In short, it's a purpose-built infrastructure platform built to deliver performance at scale.
Compared to the NVIDIA HIPPER build Aption, the NVIDIA GB200 NVL72 systems deliver 3x faster Training performance on the largest model tested by MLPERF Training Hoteldncrks benches, and almost 2x better performance per dollar.
And nvidia GB300 NVL72 delivers more than 4x speedup compared to nvidia hopper.
These performance benefits help AI developers shorten development cycles and ship new models much faster.
Testimonials on models across all Modiolities
Most of today's leading language models are trained on NVIDIA platforms.
Ai doesn't just mean text.
Nvidia supports AI development across multiple modalities, including speech, image and video generation, as well as emerging areas such as biology and robotics.
For example, models such as Evo 2 reverse gene sequences, OpenFold3 predicts 3D protein structures and boltz-2 simulates drug interactions, helping researchers to quickly find promising candidates.
On the clinical side, NVIDIA CLARA synthesis models generate realistic medical images for observational testing and diagnosis without disclosing patient data.
Companies like Runway and Inworld are riding on Nvidia's infrastructure.
Runway last week announced the Gen-4.5, a new generation frontier model that is the current top video model in the world, according to the analysis leaderboard.
Now optimized for Nvidia Blackwell, Gen-4.5 was completely developed on Nvidia GPUS throughout the research and development experience, pre-training, post-production training.
Runway also announced the GWM-1, an Nvidia Blackwell-trained virtual world model built for real-time virtual reality simulation. It is interactive, practical and general purpose, for video game, educational, scientific, entertainment and robotics applications.
Benchmarks show why.
MLERFF is a standard performance benchmark. In the latest round, Nvidia has submitted results in all seven benchmarks 3.1 benchmarks, showing strong performance and diverse performance. It was the only delivery platform in each category.
Nvidia's ability to support various AI workloads helps data centers use resources efficiently.
That's why AI Labs like black forest labs, core, mistral, Vulai, show and think about NVIDIA Blackwell machines.
Nvidia Blackwell Avoid clouds and data centers
Nvidia Blackwell is a widely available cloud service provider, Neo-Clouds and Server manufacturer.
Along with Nvidia Blackwell Ultra, which offers more performance, memory and architecture, it is now releasing server makers and cloud service makers.
Major Cloud Service Providers and Nvidia Cloud partners, including Amazon Web Services, Coreweave, Lambda, Microsoft Azure, Nebius Cloud Infrastructure powered by Nvidia Blackwell, ensures continuous operation.
From Frontier models to everyday ai, the future is built at Nvinia.
Read more about Nvidia Blackwell Platform.
