New Delhi [India], September 9 : Artificial intelligence research and operational workflows have crossed a new automation threshold, with OpenAI’s internal teams now deploying 3.1 agent-workdays of automated effort for every single workday of human labour.
According to data from OpenAI, the metric reflects an operational shift within the organization as autonomous systems manage an expanding share of development tasks alongside human personnel.
"Teams are leveraging agents to resolve infrastructure problems that once required specialist support," OpenAI stated in the update on its research operations.
The organization noted that human personnel continue to guide the overarching direction while automated workflows execute routine engineering and technical duties.
"People still set research priorities and judge results, but agents give them more capacity to pursue promising ideas," OpenAI said.
Beyond routine internal operations, the organization reported that algorithmic systems have contributed directly to advanced theoretical research. An internal model developed by the laboratory produced a solution to the Navier–Stokes Millennium Prize Problem, addressing a mathematical question that remained unresolved for roughly nine decades.
The AI organisation reported that its software products now serve more than one billion weekly active users and 2.5 million corporate enterprises across consumer, API, and enterprise deployments. 
An internal tracking analysis of individual account holders showed that daily interaction volume rose approximately 50 per cent six months after account creation compared to the first month, alongside a doubling of unique tasks attempted per user.
To manage the infrastructure demands required by continuous usage, the laboratory introduced adjustments across both computing hardware and model deployment software.
"GPT-5.6 Sol helped improve our production serving software, reducing end-to-end serving costs by 20%," OpenAI said.
It noted that additional software optimizations increased token-generation efficiency by more than 15 per cent, allowing its compute systems to generate higher output.
On the hardware side, OpenAI developed its first custom silicon accelerator, named Jalapeño, designed to handle model inference workloads. Testing on the InferenceX benchmark across three public models showed the custom processor generated 1.5 to 1.9 times the peak token throughput per watt of evaluated commercial systems, alongside a 1.7 to 3.6 times reduction in end-to-end latency.
"We plan to begin deploying it by year-end alongside accelerators from NVIDIA, AMD and other partners," OpenAI said.
The technical developments coincide with the introduction of GPT-6 Astra, an advanced model designed for complex workloads including computer navigation, code synthesis, cyber operations, and scientific analysis. 
“Years of investment in research, products, and compute are coming together. Our consumer and enterprise reach gives each model advance a direct route to customers. Our full stack compute strategy helps us deliver that intelligence with better performance and economics,” the company stated.