OpenAI reveals that researchers consume over $7,000 in tokens per day, aiming to achieve full automation by 2028
OpenAI publicly disclosed data on the scale and cost of AI programming agents used by researchers, showing that high-frequency users consume over $7,000 per day on average.
In early September 2026, OpenAI publicly disclosed the scale and cost data of AI programming agents used by its researchers. The data showed that the most active researchers consumed over $7,000 (approximately 47,000 yuan) in token value daily, and the average daily expenditure for regular researchers increased from $162 in July to $600. Over the past three months, usage of this tool has seen explosive growth, exceeding the rates of other teams within the company. AI has been deeply involved in code writing, experimental iteration, and advanced task processing, resulting in a more than half decrease in the daily number of posts by human technical support since January. OpenAI stated that it has achieved its goal of creating “autonomous research interns” and plans to establish fully automated AI researchers by March 2028. It should be noted that the above cost data is based on estimated prices from publicly available APIs and does not represent actual payment amounts.
Over $7,000Average daily Token consumption value for most active users
$162Median daily Token consumption for regular researchers in July
$600Median daily Token consumption for regular researchers in mid-August
Other teams within the companyComparison with previous three months’ figures
Key entitiesKEY ENTITIES
AmazonDave TreadwellIT HomeMetaOpenAIPeter SteinbergerSam AltmanTom Carter
Event frameEVENT FRAME
Product update
AI · 中外媒体across 1 days
Status
OpenAI publicly disclosed data on the scale and cost of AI programming agents used by researchers, showing that high-frequency users consume over $7,000 per day on average.
Coverage · reports per dayLANGUAGE SPLIT
Coverage mixSOURCE MIX
Chinese media 3English media 1
Entity relations
Integrated timelineUNIFIED TIMELINE
2026-09-07
OpenAI disclosed data on AI programming agents used by researchers
OpenAI publicly announced the scale and cost of AI programming agents used by researchers. The data shows that the most active researchers consume over $7,000 in daily Token value (approximately 47,000 yuan), while the median daily consumption for regular researchers increased from $162 in July to $600. Over the past three months…
OpenAI first disclosed the scale of usage by its researchers on AI programming agents, showing that the most frequent users consume more than $7,000 in token value per day. Last week, the company announced in a blog post that usage of this technology had increased explosively over the past three months; employees are writing code faster and conducting more experiments, with advanced tasks being gradually handed over to AI. Data shows that as of mid-August, average daily token expenditure for regular researchers exceeded $600, while the top 10% of users spent more than $7,000 per day. Since January, the number of daily posts from human technical support has decreased by more than half. OpenAI stated that it has achieved its goal of creating “automated research interns” and plans to establish fully automated AI researchers by March 2028. These costs are estimated based on publicly available API prices and do not represent actual payments.
In early September 2026, OpenAI publicly disclosed the scale and cost of AI programming agents used by researchers. Data showed that high-frequency users consumed an average of over $7,000 (approximately 47,000 yuan) in tokens per day, a significant increase from an average of $600 in July; the usage of this tool has grown faster than other internal teams over the past three months. AI has been deeply involved in code writing, experimental iteration, and troubleshooting technical issues, resulting in a halving of posts on internal support channels. OpenAI stated that it has achieved its goal of “automated research interns” and is advancing the goal of fully autonomous AI researchers by March 2028. It should be noted that costs are estimated based on API market prices and not actual expenditures.
OpenAI researchers use AI coding agents, with an average daily token consumption of over $7,000. The company revealed in its blog on Sunday that the most active researchers consume over $7,000 in tokens per day, while the median researcher consumes over $600 per day, a significant increase from $162 in July. Over the past three months, the use of AI coding agents has surged and surpassed that of other teams, with researchers using them to accelerate code delivery, expand experiments, and handle advanced tasks. Due to AI-assisted debugging, the number of daily posts on manual support channels has decreased by more than half since January. OpenAI states that it has achieved the goal set by CEO Sam Altman to build “automated research interns” capable of executing human instructions, and plans to realize fully automated AI researchers by March 2028. The data is based on estimated API推理 prices and does not represent actual payments.
OpenAI revealed that its researchers use AI to program agents, with the most frequent users consuming over $7,000 in value per day in tokens. Last week, the company posted a blog stating that usage of this technology had increased dramatically over the past three months, with the average daily token consumption for regular researchers rising from $162 to $600. AI has significantly changed how work is done: employees can deliver code more quickly, and complex tasks are completed by agents. Since January, the number of posts generated by human technical support has decreased by more than half. OpenAI has achieved its goal of creating “automated research interns” and plans to establish fully automated AI researchers by March 2028. While some companies previously stopped using internal usage rankings due to high token costs, OpenAI is positive about high consumption levels. CEO Sam Altman mentioned that the highest monthly consumption within the company reached 100 billion tokens.