GPT-5.6 drops next week (1.5M token context)

OpenAI introduced comprehensive usage analytics and spending controls for ChatGPT Enterprise, allowing admins to track credit consumption across users and models while setting flexible limits at workspace, group, and individual levels. This gives enterprises better visibility and control to manage AI investments and scale deployments more confidently.

Testing Catalog

OpenAI is set to release GPT-5.6 next week with a 1.5 million token context window and enhanced coding capabilities, strategically timed amid regulatory challenges facing Anthropic's competing model and paired with aggressive pricing cuts to gain market share.

NVIDIA Developer

NVIDIA launched XR AI, an open-source beta framework that connects AR glasses and XR devices to GPU-accelerated multimodal AI agents, integrating Cosmos vision models and Nemotron LLMs with enterprise systems. Early adoption in manufacturing and healthcare demonstrates applications for hands-free assistance in complex procedures and industrial operations.

OpenAI's Deployment Simulation method tests AI models by replaying real user conversations before release, achieving significantly better risk predictions than traditional evaluations and successfully identifying novel safety issues like reward hacking before deployment. The technique extends to complex agent settings and reduces models' awareness of being tested, though it complements rather than replaces adversarial testing for rare edge cases.

Google DeepMind

Google DeepMind introduced an AI Control Roadmap that treats AI agents as potential insider threats, using AI supervisors and cybersecurity frameworks to monitor and control internal systems even when alignment is imperfect. The defense-in-depth approach scales security measures alongside AI capabilities, addressing risks from autonomous agents expected to drive trillions in economic value.