Cursor: Partnered with SpaceXAI to train Grok 4.5, surpassing models like Opus 4.8 in multiple benchmarks Cursor announced that it has collaborated with SpaceXAI to train Grok 4.5, calling it the most powerful model to date and the first model not limited to software engineering scenarios. According to the benchmark data released, Grok 4.5 scored 83.3% on Terminal-Bench 2.1, 62.0% on DeepSWE 1.0, and 64.7% on SWE-Bench Pro, all outperforming comparison models like Opus 4.8, GPT-5.5, Composer 2.5, and Fable 5. However, in the SWE-Bench Multilingual test, it scored 78.0%, which is lower than Opus 4.8's 84.4%.