Zhipu's GLM 5.3 Now Available in Cursor with Top Coding Benchmark Performance

Z.ai has brought its latest open-weight model GLM 5.3 to Cursor, integrating the 753-billion-parameter coder into the IDE's model selection with performance that leads the open-source category on Cursor's internal benchmark. GLM 5.3 Max, the model's highest-reasoning-effort variant, scored 42.6% correctness on CursorBench 4.0, Cursor's suite of ambiguous, multi-file coding tasks drawn from real IDE sessions. The three-week-old model from Zhipu represents a milestone in open-weight AI: all gains came from scaled post-training—reinforcement learning and scheduling innovations—rather than architectural expansion, demonstrating that training methodology can match pretraining scale in advancing coding capability.