The Chinese startup’s latest AI model ranked fourth in one of the industry’s most closely followed intelligence rankings, and costs a fraction of what Anthropic or OpenAI charge for comparable performance.
GLM-5.2, released last month by Beijing-based Z.ai, has become the talk of Silicon Valley for its coding and agency capabilities that approach leading American systems, prompting comparisons with DeepSeek’s debut that shakes the market in 2025.
According Artificial analyzesGLM-5.2 scores 51 on the company’s Intelligence Index v4.1, ranking it fourth overall and first among all open weight models, ahead of the MiniMax-M3, DeepSeek V4 Pro and Kimi K2.6.
In Code Arena Front-End Coding Leaderboardthe Max level of the model comes in second, ahead of Anthropic’s Claude Opus variants.
The price is the part that has made rivals uneasy. Z.ai charges $1.40 per million input tokens and $4.40 per million output tokens on its own API, and third-party hosts list it even lower.
Various outlets have pegged that to be between one-fifth and one-seventh of what Claude Opus or GPT-5.5 cost per output token, although the exact ratio changes depending on the provider and tier being compared. Either way, it undermines the closed US border by a wide margin.
David Sacks, who was the White House AI czar under Donald Trump before returning to the private sector, said the model was “as good as currently available models from OpenAI and Anthropic”, describing it as sitting just below Anthropic’s Opus 4.8 and roughly at the level of OpenAI’s GPT-5.5.
This is a surprising statement from someone who, just a few weeks earlier, estimated the United States’ lead over Chinese laboratories at six to nine months rather than openly granting a rough parity.
Beyond the core intelligence score, GLM-5.2 also tops Artificial Analysis’ GDPval-AA v2 metric, which measures real-world agency work rather than abstract reasoning puzzles. There it scores 1,524, ahead of MiniMax-M3 and DeepSeek V4 Pro, and close enough to GPT-5.5 at its highest reasoning settings that the two are effectively tied.
Analysts have pointed out one trade-off worth noting: the model uses noticeably more output tokens per task than its open-weight peers, reducing some of its cost advantage in practice, even when the sticker price remains comparatively low.
GLM-5.2 also arrives with a detail that matters more than benchmark scores: It was trained on and runs on Chinese domestic silicon, supposedly a cluster of about 100,000 Huawei Ascend 910B processors, with no Nvidia, AMD, or Intel hardware at any stage.
This is a direct response to Washington’s chip restrictions, which have aimed to slow Chinese AI progress by cutting off access to the most advanced processors.
The model weights are released under an unrestricted MIT license, meaning any company can download, modify, and run it locally for just the cost of electricity.
Timing has helped Z.ai’s case. Washington order forcing Anthropic to suspend Fable 5 and Mythos 5 for foreign users, combined with the Trump administration’s request that OpenAI staggers the launch of its next modelhas left a gap in the market that a free, capable and downloadable alternative is well positioned to fill.
That backdrop lies within the broader story of Tightening US export controls on advanced chips.that Chinese labs have spent the last two years fixing instead of waiting.
Z.ai’s listed entity has felt the effect directly. The Hong Kong-listed company’s shares have risen sharply since its stock market debut in January, with trading spikes around the launch of GLM-5.2 sending the stock up several multiples above its initial public offering price. A successor model, the GLM-5.5, is expected in August, although Z.ai has not confirmed a firm date.
