Chinese AI labs are copying OpenAI's newest model by studying its answers

Some of China's leading AI labs are studying OpenAI's newest model to build their own. According to The Information, DeepSeek and Moonshot AI relied heavily on answers produced by GPT-6 Astra, OpenAI's most powerful model, which came out in early September.

What is happening

The practice is called distillation. A new AI is trained on the answers of a stronger AI, so it picks up many of the same skills. Think of a student copying a top performer's worked solutions instead of learning from textbooks.

The Information reports that Astra has met with strong demand since its release. One growing group of customers is Chinese AI labs looking to pull out its capabilities.

Why it matters to you

If rivals can copy the skills of a top model, they can offer similar tools at lower cost. That may change what you pay for AI and which tools you pick for your business or home.

• Rival AI tools could get closer in quality to the leading ones.
• Competition could put pressure on prices.
• The choice between AI tools may get harder.

What is not known yet

The full article was not available, so key details are unclear. The material does not say how much of each lab's work depended on Astra, how the outputs were gathered, or how OpenAI plans to respond. It also does not say whether the resulting tools match Astra in quality.

What to watch

Watch for new releases from DeepSeek and Moonshot AI, and for any statement from OpenAI on the matter. Price changes across AI tools may be the first sign of any effect on everyday users.