OpenAI has released GPT-5.2, an impressive upgrade over its predecessor GPT-5.1, showcasing significant advancements in various benchmarks, demos, and real-world applications. The video highlights visually stunning demos such as Flavio Adamo’s 3D realistic bouncing balls in a hexagon and Ethan Mollik’s infinite Neo-Gothic city shader with realistic water physics. These demos demonstrate GPT-5.2’s enhanced ability to generate complex, visually rich content with improved physics and lighting effects, setting a new standard for AI-generated visuals.

Benchmark results reveal GPT-5.2’s state-of-the-art performance across multiple challenging tests. It achieved a 5% improvement on the Swebench Pro benchmark and a 4% increase on the GPQA Diamond science benchmark. Remarkably, GPT-5.2 aced the difficult Amy 2025 math competition with a perfect score, outperforming competitors like Gemini 3 Pro and Claude Opus 4.5. The most striking improvement was on the ARC AGI 2 benchmark, which tests learning and generalization abilities, where GPT-5.2 jumped from 17% to nearly 53%, establishing itself as the leader in AGI-related tasks with dramatically improved cost efficiency.

GPT-5.2 also excels in practical, economically valuable tasks such as workforce planning, cap table management, and project reporting. Compared to GPT-5.1, it produces more accurate and better-formatted Excel sheets, correctly calculating complex financial formulas that GPT-5.1 struggled with. Its ability to generate clear, visually appealing reports and handle sophisticated coding tasks, like creating a realistic ocean wave simulation app with adjustable parameters, highlights its versatility and reliability for professional use cases.

In addition to improved accuracy and reasoning, GPT-5.2 shows significant advancements in visual reasoning and tool use. It cuts error rates in chart reasoning and software interface understanding by roughly half and demonstrates superior performance in identifying components in images, such as motherboard parts. Its tool-calling capabilities have nearly doubled, enabling it to handle complex multi-step interactions more effectively. Furthermore, GPT-5.2 exhibits safer behavior with fewer hallucinations and improved mental health evaluation capabilities, making it a more trustworthy assistant.

Despite these impressive improvements, GPT-5.2 comes with a higher price point, reflecting its enhanced capabilities. Input token costs increased from $1.25 to $1.75 per million, and output token costs rose from $10 to $14 per million. Nevertheless, the efficiency gains, especially in AGI benchmarks, justify the cost for many users. The video also mentions ongoing community engagement through prediction markets and a major giveaway, encouraging viewers to explore GPT-5.2’s capabilities and stay tuned for future updates as OpenAI continues to push the boundaries of AI performance.