Skip to content
NewA calmer, faster ZeroHour

OpenAI says internal coding agents now outpace researcher pay

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

The Decoderby Maximilian Schreiner ()3 min read1
AI summary · x-ai/grok-4.7Summaries by Vani AI

At list prices, median October token use was $2,400 a day, while agents still fail often on harder engineering work.

OpenAI’s second internal research-acceleration report says use of coding agents such as Codex and Dots is rising as the company tracks signs of recursive self-improvement. At regular API list prices, the median researcher’s October token use was worth $2,400 a day, up 62 percent month over month, and $23,000 a day for the top 10 percent. Success on unattended multi-hour tasks improved but stays low on complex engineering. Staff surveys prefer keeping the agents over large cuts in people or compute, while interviewees say models lack research taste and need human oversight.

Unconfirmed: Token figures are theoretical list-price equivalents, not cash spend, and exclude equity that OpenAI says is 70 to 80 percent of pay; the surveys are internal and non-representative.

Key facts
  • Agents equated to 5.7 workdays per researcher per calendar day, about eight in pre- and post-training.
  • Unattended success rose from 23 to 46 percent on two- to four-hour tasks between March and October.
  • Training-run monitoring succeeded 79 percent of the time; inference and serving code, 25 percent.
  • A survey of 101 employees had a median preference to drop 40 percent of staff rather than lose agents for three months.
  • Key ideas for GPT-6 Astra and GPT-6.2 Astra still came from humans, except some pretraining insights via Codex.
Explain this AI news with Vani AIWhat was announced, what is new and why it matters, in a new chat

From the original article

3 min read at the-decoder.com

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. The article OpenAI's own researchers now say they value their AI tools more than nearly half their colleagues appeared first on The Decoder.

The Decoder

By Maximilian Schreiner · the-decoder.com

Read the full article at The Decoder

A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.