SAN FRANCISCO — xAI is raising the bar again. The company has begun rolling out Grok 5, its most capable artificial-intelligence model to date, trained on the vast Colossus 2 supercomputer and tuned for sharper reasoning, longer memory and noticeably faster answers.
The launch cements a breakneck release cadence for the young lab, which has gone from a standing start to shipping frontier models in a matter of years. For users, the upgrade promises a Grok that thinks more carefully, remembers more of a conversation and handles complex, multi-step problems with fewer stumbles.
Trained on Colossus 2
Grok 5's biggest advantage may be the machine behind it. xAI trained the model on Colossus 2, the Memphis supercomputer the company has been aggressively scaling. The lab recently moved to double the Nvidia GPU count powering Colossus 2 on its way toward a million-chip target, giving Grok 5 far more training compute than its predecessors.
More compute, applied well, tends to translate into better models — stronger reasoning, broader knowledge and more reliable behavior. xAI says Grok 5 posts significant gains on math, coding and reasoning benchmarks, and early testers describe it as quicker and more consistent than the version it replaces.
What Users Get
The headline improvements are practical. Grok 5 carries a larger context window, letting it hold long documents and extended chats in mind without losing the thread. Its reasoning is more deliberate on hard problems, and its responses come back faster thanks to inference optimizations that ride on the expanded cluster.
