Cursor adds Z.ai's GLM 5.3 and says it tops open models
Cursor says GLM 5.3 Max is the best-scoring open-weight model on its own CursorBench 4.0, and has added both GLM 5.3 and a Flash variant to its editor.
Cursor said on X on Thursday evening that GLM 5.3 and GLM 5.3 Flash are now available in its coding editor. In the same post the company said GLM 5.3 Max is the best-scoring open-weight model on CursorBench 4.0, a benchmark Cursor runs itself. The post had about 900 likes when we read it.
https://x.com/cursor_ai/status/2105787358557999585
The scoring claim is Cursor's alone. We have not seen the CursorBench 4.0 methodology or the scores, and no outside group has published a comparison. Cursor's post does not say what a Max tier is or how it differs from the base model. Z.ai, the Chinese lab behind GLM, had not commented in the material we reviewed.
GLM-5.3 itself dates from August. According to an NYU Shanghai research note, Z.ai announced it on 14 August as a 743-billion-parameter mixture-of-experts model on the same base as June's GLM-5.2, with every gain coming from longer post-training. Z.ai held the weights back for roughly two weeks for a safety review.
The NYU note reports Z.ai's own coding numbers: Terminal-Bench rose from 4.6 to 28.3 and DeepSWE from 46.2 to 66.9 against GLM-5.2. It also says the model found 2,436 vulnerabilities across 269 open-source projects, 1,097 of them rated critical or high. Those figures come from Z.ai and have not been independently checked.
Gigazine reported that the weights went public on 29 August under a bespoke GLM-5.3 licence that requires a Z.ai security review for companies with more than $10 billion in annual revenue. The licence terms matter for anyone planning to ship the model inside a product, and they should be read directly.