Grok 4 Heavy
xAI · 2025-07-09
Grok 4 Heavy is the parallel test-time-compute tier of xAI's Grok 4, launched in July 2025 for SuperGrok Heavy subscribers; it runs several Grok 4 agents on a problem at once and reconciles their hypotheses. xAI reports its benchmark results under the Grok 4 Heavy name separately from Grok 4; it has no separate API model identifier.
Benchmark scores
| Benchmark | Score |
|---|---|
| Humanity's Last Exam (with tools) | 44.4 |