On July 14, Sam Altman, CEO of OpenAI, posted on the X platform, showcasing impressive performance metrics of its AI agent products (including Codex and ChatGPT Work) - Token usage increased by 2.5 times last week. This growth was directly driven by the official launch of the GPT-5.6 series model in ChatGPT, Codex, and API on July 10, which significantly boosted the usage of AI agent products.

ChatGPT

Improved programming capabilities, 54% higher token efficiency

In an interview with CNBC, Altman stated that the GPT-5.6 Sol model performed "as well as, or even better than" the mainstream competitive models in AI agent programming tasks, while achieving a 54% improvement in token efficiency. This means users can not only get higher quality outputs but also complete tasks at a lower cost, significantly improving the cost-effectiveness.

Internal usage data is equally impressive. OpenAI previously reported that by June this year, 98% of employees were using Codex. In seven months, Codex's application in research areas grew by 56 times, and the demand for 8-hour autonomous tasks increased by 10 times. These figures indirectly confirm that AI agents are moving from a trial phase to an essential part of work processes, and the release of GPT-5.6 has further fueled this trend.