SpaceXAI Launches Grok 4.5, Narrows Gap with GPT 5.5 and Opus 4.8

SpaceXAI has launched Grok 4.5, its latest AI model, positioning it for coding, agentic tasks, and knowledge work. 

The model is available in Grok Build, Cursor, and the SpaceXAI API. However, it is not yet available in the European Union and is expected to be available there in mid-July.

“Grok 4.5 is SpaceXAI’s smartest model built to excel at coding, agentic tasks, and knowledge work,” the company said. “It’s our strongest model ever and was trained alongside Cursor.”

Grok 4.5 runs at 80 tokens per second. It can handle software engineering tasks ranging from Rust and C/C++ programming to building complete applications from a single prompt.

SpaceXAI has priced Grok 4.5 at $2 per million input tokens and $6 per million output tokens. 

Grok 4.5 is a mixture-of-experts (MoE) model. Cursor said the training included trillions of tokens from Cursor, covering developer interactions with codebases and software tools. Grok 4.5 was trained on datasets covering coding, science, engineering, and mathematics, with reinforcement learning focused on software engineering and other technical tasks. 

According to SpaceXAI, the model was trained across tens of thousands of NVIDIA GB300 GPUs using data filtering, deduplication, quality scoring, and domain-specific selection.

On software engineering benchmarks, SpaceXAI said Grok 4.5 achieved a 62.0% pass@1 score on DeepSWE 1.0, behind Fable Max at 66.1% and GPT 5.5 xhigh at 64.31%. 

On SWE Marathon, however, it led with a 29.0% resolution rate, ahead of Opus 4.8 Max at 26.0% and Fable Max at 24.0%. On Terminal Bench 2.1, it scored 83.3%, close to Fable Max’s 84.3% and GPT 5.5 xhigh’s 83.4%.

SpaceXAI also said the model generates fewer output tokens than competing models. On SWE Bench Pro, Grok 4.5 used an average of 15,954 output tokens per task, compared with 67,020 for Opus 4.8 Max, which the company said translates to about 4.2 times fewer tokens.

The company said the model delivers “roughly 2x the token efficiency of comparable leading models,” reducing the number of steps required to complete tasks.

It is also the default model in Grok Build, where it can create Excel models involving web research, multi-sheet formulas, and notes, as well as PowerPoint presentations and Word documents.