Chinese AI Models Close the Gap With Anthropic and OpenAI
The New York Times ● Covered by 3 sources
Chinese lab Z.ai dropped GLM-5.2, an open source model that nearly matches Anthropic's top offerings. It's cheaper, unrestricted, and already cracked the global top 10 for open models.
Based on reporting by The New York Times — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Z.ai just put out GLM-5.2, and the gap between Chinese and American frontier AI just got noticeably smaller. The model reportedly performs close to Anthropic's most capable, export-restricted systems, but without the licensing hurdles and price tags that come with US-based providers. For a market that's spent two years watching American labs set the pace, that's a meaningful shift.
What makes this notable isn't just the benchmark scores. It's the distribution. GLM-5.2 is open source, and it has already climbed into the top 10 most popular open models worldwide, according to TLDR's reporting. That kind of adoption curve matters more than a single leaderboard result, because it signals developers are actually building on it, not just admiring it in a press release.
Cost is the other half of the story. Z.ai is offering near-frontier performance at a fraction of what Anthropic or OpenAI charge for comparable capability, and it's doing so without the US export restrictions that keep the most advanced American models locked behind enterprise contracts and government scrutiny. For startups and researchers outside the US, particularly in regions where access to restricted models is legally or practically difficult, that combination of cheap and open is hard to ignore.
None of this means Anthropic or OpenAI are suddenly behind. Their most advanced systems still likely outperform GLM-5.2 on the hardest tasks. But the margin that once looked comfortable is shrinking fast, and it's shrinking in public, on open leaderboards, where anyone can check the claims themselves. That's a different kind of competitive pressure than a rival simply saying they're catching up.
My take — AI-written commentary, not fact-checked reporting
I've said it before: the moat American AI labs are counting on was always thinner than they wanted to admit, and GLM-5.2 is more proof. Export controls slow things down, they don't stop them, and cheap open models winning real developer adoption is exactly the pattern that should worry anyone betting on closed models staying ahead by default.
Read more about this at: The New York Times