SpaceXAI launches Grok 4.5 for coding and agentic work
The story in five sentences
- SpaceXAI has introduced Grok 4.5 for coding, agentic tasks, and knowledge work.
- The company says the model was trained in collaboration with Cursor.
- The vendor reports an output speed of 80 tokens per second.
- API access costs two dollars per million input tokens and six dollars per million output tokens.
- Performance and efficiency comparisons come from SpaceXAI and have not been independently verified.
Full explanation
SpaceXAI introduced Grok 4.5 on July 16 for programming, science, engineering, mathematics, agentic tasks, and knowledge work. The company says the model was trained with Cursor. It is available through Grok Build, Cursor, and the provider’s console. The reported output rate of 80 tokens per second is a vendor claim.
The announcement does not show Grok 4.5 leading every cited benchmark. Other systems rank ahead on Grok DeepSWE, while SpaceXAI reports 29 percent on SWE Marathon. The company also contrasts 15,954 generated output tokens with 67,020 for Opus 4.8 in one comparison. That selection and its harness do not support a neutral ranking across models.
API pricing is two dollars per million input tokens and six dollars per million output tokens. The model is described as temporarily free, but no end date is stated. SpaceXAI promotes Grok 4.5 for office work in addition to coding and agentic uses. Claims about performance, speed, and token efficiency remain provider evaluations and have not been independently confirmed.