Grok 4 Heavy

Grok 4 Heavy launched on July 9, 2025 alongside Grok 4 as xAI's premium reasoning tier. Where standard Grok 4 is a single-instance reasoning system, Heavy spawns up to 16 parallel instances of Grok 4 that each work through the same query independently, then compare outputs and converge through a debate-style synthesis, like a study group comparing notes.

The approach produced headline benchmark numbers: 100% on AIME 2025, 96.7% on HMMT 2025, 88.4% on GPQA Diamond, and 44.4% on Humanity's Last Exam with tools enabled, the first publicly available model to cross 40% on that benchmark. It is sold through the $300/month SuperGrok Heavy subscription and the xAI API.

The honest caveats: sixteen parallel inference passes cost substantially more compute than a single pass, which is exactly why it carries premium pricing, and all benchmark figures are xAI-reported. There are no open weights.