Grok 4.5 arrives as SpaceXAI’s most capable model for coding and agents

SpaceXAI has launched Grok 4.5, its most advanced model to date, built for coding, agentic tasks and knowledge work. The company says it was trained alongside Cursor and describes it as its strongest system so far.

The model was trained on datasets spanning coding, science, engineering and math, across tens of thousands of Nvidia GB300 GPUs, using domain-focused data filtering and selection techniques. Reinforcement learning was applied to hundreds of thousands of software engineering tasks, with automated and model-based grading. According to the company, its infrastructure allows for highly asynchronous training, where agentic processes run for hours while learning continues in parallel across tens of thousands of GPUs.

On the DeepSWE benchmark (pass@1), Grok 4.5 scores 62.0%, below Fable max's 66.1% and GPT-5.5 xhigh's 64.31%, but ahead of Opus 4.8 max's 55.75% and Opus 4.7 max's 40.12%.

SpaceXAI also highlights response speed, at 80 tokens per second, and efficiency: on SWE Bench Pro tasks, Grok 4.5 uses an average of 15,954 output tokens, compared to 67,020 for Opus 4.8 max — 4.2 times fewer, which the company says cuts both time and cost per task.

The model is now the default in Grok Build, the company's tool for office content: it can build Excel models with complex formulas and web research, design slides using native PowerPoint shapes, and write Word documents.

Access is priced at $2 per million input tokens and $6 per million output tokens. Grok 4.5 is available today in Grok Build, on all Cursor plans, and through the SpaceXAI console.

Key points

• SpaceXAI launches Grok 4.5, its most advanced model to date
• Trained alongside Cursor, built for coding, agents and knowledge work
• Trained on tens of thousands of Nvidia GB300 GPUs with asynchronous reinforcement learning
• Scores 62.0% on DeepSWE, ahead of Opus 4.8 max (55.75%) but behind Fable max (66.1%) and GPT-5.5 xhigh (64.31%)
• Runs at 80 tokens per second and uses 4.2x fewer tokens than Opus 4.8 max on SWE Bench Pro
• Now the default model in Grok Build, for Excel, PowerPoint and Word
• Costs $2 per million input tokens and $6 per million output tokens
• Available in Grok Build, in Cursor, and through the SpaceXAI console