OpenAI has announced significant advancements in the price-performance frontier of its GPT-5.6 models. Specifically, GPT-5.6 Luna, their fastest and most affordable model, is now 80% cheaper, while GPT-5.6 Terra, a balanced model for everyday work, sees a 20% price reduction, enhancing the return on investment for enterprises using AI.

A new Fast mode has also been introduced for GPT-5.6 Sol, offering API customers up to 2.5 times faster speeds compared to standard processing. These improvements are attributed to enhanced model efficiency, optimized inference systems, and smarter agentic harnesses that connect models to tools and context.

Notably, GPT-5.6 Sol has played a crucial role in these efficiencies by autonomously rewriting and optimizing production kernels, designing experiments to improve token generation, and monitoring training processes. This initiative is central to OpenAI's mission of ensuring AGI benefits all of humanity by making frontier-level intelligence more abundant and affordable.

👉 Read the full article on OpenAI

This arXiv paper presents a mathematical proof demonstrating that Maxwell's conjecture is false. Maxwell's conjecture posited that the electrostatic potential of n point charges has at most (n-1)^2 non-degenerate critical points.

The authors exhibit a configuration of five point charges in Euclidean space whose electrostatic potential admits at least 24 non-degenerate critical points, which exceeds the (n-1)^2 upper bound (16 for n=5) proposed by Maxwell.

This finding is significant in classical physics and mathematical physics, challenging a long-standing conjecture. While the title includes "GPT 5.6 Sol," the abstract and content of the paper itself do not explicitly state that a specific AI model was used to derive this proof.

👉 Read the full article on arXiv

Google's Gemini Spark is expanding its capabilities with new Chrome web browsing integration and broader global access. With user permission, Spark can now handle complex "web errands" using logged-in accounts and saved passwords, such as scheduling apartment viewings or researching flight options and initiating bookings.

This AI agent functionality is designed to automate daily tasks while incorporating safeguards against threats like prompt injection and requiring user confirmation for sensitive actions such as payments. The feature is initially rolling out in the U.S. with plans for expansion to additional regions in the future.

Furthermore, access to Spark is being extended to Google AI Pro subscribers in over 160 additional countries starting today, allowing more people to begin automating their daily tasks.

👉 Read the full article on Google Blog