NVIDIA AI Updates: April 24, 2026

1. NVIDIA Reports GPT-5.5 Codex Is in the Hands of 10,000 of Its Own Employees

NVIDIA. On launch day, NVIDIA said over 10,000 of its employees are using GPT-5.5-powered Codex internally to accelerate workflows, running on NVIDIA’s GB200 NVL72 systems. NVIDIA claims the deployment delivers 35x lower cost per million tokens and 50x higher token output per second per megawatt compared to prior generations — figures that are as much a sales pitch for GB200 NVL72 as they are a Codex endorsement. The post underscores a decade-long partnership with OpenAI and positions NVIDIA as the reference platform for GPT-5.5 scale deployments. Source

2. A Three-Agent LLM Swarm Won a Kaggle Competition With 600,000 Lines of Code

NVIDIA. NVIDIA’s developer blog walks through how three coordinated LLM agents generated more than 600,000 lines of code, ran 850 experiments, and took first place in a Kaggle playground competition. The post is less about the competition result and more about a reproducible pattern: an orchestrator agent, a feature-engineering agent, and a modeling agent running on GPU-accelerated infrastructure so each iteration takes minutes instead of hours. It is a useful concrete template for teams building their own ML research agents. Source