CHINA’S OPEN-WEIGHT GLM-5.2 BEATS GPT-5.5 ON CODING BENCHMARKS FOR 96 PERCENT LESS MONEY GLM-5.2 from Zhipu AI, released under an MIT license in mid-June, posted a SWE-bench Pro coding score of 62.1 against GPT-5.5’s 58.6 — while costing $4.40 per million output tokens versus OpenAI’s $30 per million. The math is worth sitting with. A company using GPT-5.5 for code-heavy workloads is paying roughly seven times more per token for a model that now scores lower on the benchmark that matters most to software development shops. That is not a marginal efficiency gain. For high-volume users, the cost differential alone could justify switching vendors without ever running an internal evaluation. GLM-5.2 is not the only Chinese open-weight model punching above its weight class, but it is the most recent example of a trend that has been accelerating throughout 2026. Models coming out of Chinese labs are increasingly competitive on both performance and pricing benchmarks, and they are releasing under open licenses that allow organizations to run them on their own infrastructure without any per-token fees at all. The pattern here should be concerning for the major Western AI labs. They are building their business models around proprietary API access at premium prices. Open-weight models do not compete on the same terms. They compete by making the business model itself unnecessary. Every benchmark cycle that a Chinese open-weight model matches or exceeds a frontier closed model narrows the justification for paying frontier prices. Keywords: GLM-5.2 Zhipu AI, Chinese AI model, open-weight AI benchmark, AI pricing competition