Zhipu GLM-5.3 Lands in Cursor With a 42.6% CursorBench Score, and the Stock Jumps 6%

Zhipu's GLM-5.3 has picked up the kind of endorsement that matters more than another benchmark press release: it is now available inside Cursor, one of the most widely used AI coding tools, and the company's Hong Kong-listed shares jumped more than 6 percent on October 5 to close at HK$665 after intraday gains past HK$666.

The move itself is straightforward. Cursor announced it has added GLM-5.3 and GLM-5.3-Flash to its platform, and said that under its Max reasoning setting GLM-5.3 achieved the highest score of any open-weight model on CursorBench 4.0, the coding evaluation Cursor built and maintains, at 42.6 percent. That phrasing is important: CursorBench is Cursor's own benchmark, so the number reflects performance in the vendor's own harness and workload mix rather than an independent audit.

The adoption is the latest in a string of external validations for the GLM family. Aleph Alpha, the German sovereign-AI startup, disclosed in the technical report accompanying its new Kolibri model that GLM-5.2, GLM-5.3 and Alibaba's Qwen3.8-27B were the main models used to produce its supervised fine-tuning data — meaning a European government-oriented lab is openly relying on Chinese open-weight models as teacher models. And Anthropic's September 29 frontier red team report, citing evaluations by NIST's CAISI, listed GLM-5.3 as the strongest open-weight model on cyber capabilities at the time of assessment.

Investors read the accumulation the same way. Goldman Sachs upgraded Zhipu from Neutral to Buy and set a 12-month DCF-based target price of HK$1,560, citing a clearer commercialization path. Southbound funds net bought roughly HK$7.88 billion of the stock in September alone, and the shares have been re-rated steadily since their Hong Kong listing as the open-weight business model stopped looking like a giveaway and started looking like distribution.

Some caution is warranted on the Cursor specifics. Third-party write-ups note that Cursor's technical documentation showed entries for GLM 5.2 alongside standardized pricing, and that immediate availability, subscription pricing and regional access terms for GLM-5.3 were not independently verified at the time of writing. Integrations of this kind also tend to roll out by plan tier before they reach everyone.

The strategic significance is bigger than one benchmark score. Coding is the highest-willingness-to-pay AI workload, and the tools that mediate it — Cursor among them — effectively decide which models developers touch. An open-weight Chinese model earning default placement there competes with closed frontier models on the same surface, at open-weight pricing, and that is precisely the combination enterprise buyers have been waiting for.

For Zhipu, the sequence — open weights, external adoption, then analyst upgrade — is the playbook it has been running all year, and the stock's reaction suggests the market now prices GLM distribution news as fundamental rather than promotional. The open question is durability: benchmark leads among open-weight models have historically lasted quarters, not years, and Cursor's bench will keep re-sorting the field as new releases land.