SpaceXAI has launched Grok 4.5, the latest version of its AI model, with a focus on software development, AI agents, and technical knowledge tasks. The company says it is the most capable Grok model released so far.

The new model is now available through Grok Build, Cursor, and the SpaceXAI API. Users in the European Union will have to wait a little longer, as SpaceXAI expects to roll it out there by mid-July.

Grok 4.5 is designed to help developers tackle everything from writing Rust and C/C++ code to building complete applications from a single prompt. It generates text at a speed of 80 tokens per second, making it suitable for demanding coding workflows.

For API users, SpaceXAI has set the pricing at $2 per million input tokens and $6 per million output tokens.

The model uses a mixture-of-experts (MoE) architecture and was trained on trillions of tokens, including developer interactions collected through Cursor. Its training covered coding, science, engineering, and mathematics, while reinforcement learning was used to improve its software engineering abilities.

SpaceXAI said Grok 4.5 was trained on tens of thousands of NVIDIA GB300 GPUs using a carefully selected dataset that went through filtering, deduplication, quality checks, and domain-specific data selection.

The company also shared benchmark results showing the model’s performance. On the DeepSWE 1.0 benchmark, Grok 4.5 achieved a 62.0% pass@1 score, ranking behind Fable Max and GPT 5.5 xhigh. However, it topped the SWE Marathon benchmark with a 29.0% resolution rate, outperforming both Opus 4.8 Max and Fable Max. It also scored 83.3% on Terminal Bench 2.1, putting it close to the highest-performing models in that test.

SpaceXAI says Grok 4.5 is also more efficient than many competing AI models. On SWE Bench Pro, it generated an average of 15,954 output tokens per task, compared with 67,020 tokens for Opus 4.8 Max. According to the company, this means Grok 4.5 can solve tasks with far fewer generated tokens while maintaining strong performance.

The company estimates that Grok 4.5 delivers roughly twice the token efficiency of other leading AI models, allowing it to complete complex tasks with fewer reasoning steps.

Beyond coding, Grok 4.5 is the default model in Grok Build, where users can create Excel spreadsheets with web research and multi-sheet formulas, generate PowerPoint presentations, and draft Microsoft Word documents from simple prompts.

Also Read: SpaceX Acquires xAI in Deal Valuing Combined Company at $1.25 Trillion