OpenAI: Our agents have sped up internal research by 6 months

OpenAI has released its first ever transparency report detailing how it’s using coding agents to speed up research.

In the report, OpenAI details its use of agents to automate experiments, write code, and analyze data. According to one of its product leads, once they started using Astra internally, their productivity went through the roof, allowing them to accelerate some projects by six months. Meanwhile, Anthropic says that Claude is now responsible for over 80% of its production code.

1Why This Matters for the Industry

It’s the first time a top AI lab has opened up about how it’s using AI to speed up its own R&D. By iterating faster than humans, agents shorten the time between conception and deployment of new models, enabling AI to start contributing to the development of future generations of AI. This will dramatically alter the rate of change across the field.

There are questions around the reliability and reproducibility of results generated by agents working autonomously, but the fact that OpenAI has chosen to publish internal stats suggests it sees this as a major milestone worth sharing with the world.