SpaceXAI launches Grok 4.5 with 62% DeepSWE 1.0 score
July 8, 2026
Grok 4.5 is optimized for coding and agentic workflows, achieving a 62% pass@1 on the DeepSWE 1.0 benchmark. The model outperformed Opus 4.8 and rivals GPT 5.5 in engineering and math-heavy tasks.
HOW THIS AFFECTS YOU
●
builderYou can leverage this model for specialized coding and agentic engineering workflows.
●
researcherThe training datasets specifically emphasize science, engineering, and math for improved reasoning.