Elon Musk has reiterated a rather shocking tidbit about xAI’s impressive scores at the ARC-AGI benchmark.

He also announced that xAI’s upcoming model, Grok 5, will begin training very soon.

xAI results on the ARC-AGI benchmark

In a recent post on social media platform X, Greg Kamradt, President of the ARC Prize Foundation, noted that Grok 4 has been used in two SOTA submissions to the ARC-AGI V1 and V2 benchmarks. The submissions, which were the best that the foundation has encountered to date, were done using xAI’s large language model, Grok 4.

As per Kamradt, when asked about why the submissions were done using Grok 4, the authors stated that “It was the best model I used in testing.” Musk was evidently proud of this accomplishment, though he also highlighted that these accomplishmen

See Full Page