OpenAI agent token usage jumps 2.5x in a week

OpenAI CEO Sam Altman shared that token usage for their agent products (including Codex and ChatGPT Work) jumped 2.5 times last week. This boost came after the launch of GPT-5.6 series models on July 10 across ChatGPT, Codex, and API. In a CNBC interview, Altman said the GPT-5.6 Sol model performs as well as or better than competitors in coding tasks, while being 54% more token-efficient. OpenAI also reported that as of June, 98% of its employees use Codex agents, with research use growing 56x in seven months and demand for 8-hour autonomous tasks rising 10x.