Grok-4 Heavy

Benchmarking a Multi-Agent Powerhouse

Grok-4 "Heavy Mode" emerges as a powerhouse, leveraging the parallel processing of up to 32 agents per request to achieve impressive performance benchmarks. This beefed-up version of Grok-4 excels in complex reasoning tasks, demonstrated by a 58% solve rate on Humanity's Last Exam, a significant leap over GPT-4o. While boasting an output speed of 14 tok…