AIsaAIsaStudio English 打开 AIsa Studio

OpenAI 内部报告:研究员理论 token 消耗超薪水,Agent 相当于每天 5.7 个工作日

原标题:OpenAI's own researchers now say they value their AI tools more than nearly half their colleagues

OpenAI 发布第二份内部研究加速报告,Codex 等内部 Agent 使用量激增,10 月中位数研究员的理论 token 消耗达每天 $2,400,环比增长 62%,Agent 运行相当于每位研究员每天 5.7 个工作日。自主完成 2-4 小时任务的成功率从 23% 升至 46%,但模型在复杂工程上常产出低质 "AI slop",101 名员工的调查中位数答案是宁可放弃 40% 的团队成员也不愿三个月没有编程 Agent。

为什么值得看

报道给出 OpenAI 内部 Agent 用量、成功率与员工调查等一手数据,可帮助读者评估 AI 加速研究的真实水平与短板。

阅读原文 ↗

本页的摘要、推荐理由与解读出自 AIsa Studio 的 AI 风向智能体,原文版权归原作者所有。

原文

OpenAI's internal use of coding agents is growing fast, with theoretical token costs now exceeding researchers' base salaries. Even though teams are reluctant to work without these tools, the models still produce a lot of "AI slop" beyond routine tasks.

In a second report on internal research acceleration, OpenAI has published its own data on how deeply AI agents are embedded in the development of new models. The company says it aims to document indicators of so-called recursive self-improvement, or RSI, where AI systems speed up the development of future generations.

Usage of internal tools like Codex and Dots has surged recently, according to the report. At regular API list prices, the median researcher would have consumed $2,400 worth of tokens per day in October, up 62 percent from the previous month. Among the top 10 percent of users, that theoretical value doubled to $23,000 per day.

Converted into eight-hour workdays, the agents reportedly run the equivalent of 5.7 workdays per researcher per calendar day, and about eight in the pre- and post-training teams. More than two-thirds of tracked researchers ran at least four agents in parallel at peak usage.

The theoretical API equivalent of that agent runtime was twice as high as researchers' regular salary costs in September, according to OpenAI. That calculation excludes equity packages, though, which make up 70 to 80 percent of total compensation at OpenAI. Measured against the company's total research compute, internal agent use accounted for just 2 percent.

A large share of AI usage goes toward communication and coding. | Image: OpenAI

High success rates on routine tasks, weaknesses in code scaffolding

Between March and October, the estimated success rate of autonomous agents working without human intervention on two- to four-hour tasks rose from 23 to 46 percent, according to OpenAI. For four- to eight-hour tasks, the rate climbed from 13 to 35 percent. A breakdown using an Epoch AI taxonomy shows, however, that the systems mainly handle simple tasks.

Agents reportedly complete training-run monitoring 79 percent of the time and answer technical questions independently 72 percent of the time, but they frequently fail at more complex engineering work. The success rate for building evaluation and reinforcement-learning environments sits at 29 percent, and for writing code for inference and serving systems at 25 percent.

In an internal, non-representative survey of 101 employees, researchers said at the median that they would rather give up 40 percent of their team's staff or 30 percent of their compute capacity than go three months without current coding agents. In a separate quick poll of 28 employees, 27 preferred the latest internal model over a mid-career human engineer.

Lack of research taste requires human oversight

Qualitative interviews with ten researchers revealed the downside of heavy usage. The models lack any "taste" for good research ideas and for code and data quality, according to the interviewees. Agents get lost in unnecessary steps, make wrong assumptions in analyses, and submit subpar "AI slop" code changes along with unusable data artifacts. Close human oversight is essential to avoid wasted compute cycles.

The key conceptual ideas for GPT-6 Astra and GPT-6.2 Astra came exclusively from humans in the post-training and safety areas, according to the teams interviewed. Only in pretraining for GPT-6.2 Astra did individual researchers say they gained important insights through dialogue with Codex.

OpenAI cautions in its conclusion that the data gathered so far from human-AI collaboration may not extrapolate linearly. The company points to recent, rapid AI breakthroughs in mathematics and considers it possible that similarly abrupt advances could occur in deep-learning research itself as models gain more autonomy.

The evaluation organization METR had recently estimated, after an eleven-week look inside competitor Anthropic, that AI use there had nearly doubled the pace of development. Before that, OpenAI had declared in its first research-acceleration report in early September that it had reached its self-imposed goal of an automated AI research intern. Chief scientist Jakub Pachocki warned of losing control in an accompanying post, after an audit revealed numerous gaps in the oversight system.

原文版权归 The Decoder 及原作者所有,本站转载仅供阅读。如需删除,请联系 support@aisa.one。

事件综述

OpenAI 发布内部报告:AI 深度改变内部研究

2026年11月18日,OpenAI 发布第二份内部研究加速报告,介绍 AI 如何改变其内部研究工作。报告称研究人员对 AI 的使用持续呈指数级上升:其中位数研究员 10 月的理论 token 消耗达每天约 $2,400,环比增长 62%,智能体运行量相当于每位研究员每天 5.7 个工作日。关于编码智能体的自主能力,报告称其自主完成需人类 2 到 4 小时任务的成功率已从 23% 升至 46%,较今年 3 月实现翻倍,即能独立完成近半数此类任务。但报告同时指出,模型在复杂工程任务上常产出低质量的"AI slop",人类参与仍是核心。在安全方面,报告称研究人员投入安全与对齐领域的人力增速快于其他任何研究方向,并呼吁(主张)AI 实验室应被要求跟踪并公开报告递归自我改进(RSI)相关指标。另有 X 上对报告的转引称,第 90 百分位的 OpenAI 研究员每天使用约 $23,000 的 token,较 9 月增长 105%(早先报道称这一数字,后续 The Decoder 报道则给出中位研究员每天 $2,400 的数字,两者对象不同,分别为第 90 百分位与中位数研究员)。报道还提到,对 101 名员工的调查中,受访者中位数回答是宁可放弃 40% 的团队成员,也不愿三个月没有编程 Agent。

其他来源的报道 3