— Chatbots / LLMs Tool

Grok 4 Heavy

Last updated 2026-08-04 · Reviewed by ToolForge Editorial

Multi-agent parallel reasoning from xAI's flagship tier

★ 4.4 · Undisclosed · Since 2025 (xAI) · Basic Grok 4 free; Heavy requires subscription

$300 $300/mo SuperGrok Heavy

Try Grok 4 Heavy → Read full review

Multi-agent test-time compute for hard problems

Grok 4 Heavy is the highest tier of xAI's Grok 4 family. Unlike the standard Grok 4, Heavy runs multiple internal agents in parallel — each working the problem from a different angle — then compares, synthesizes, and votes internally before answering. xAI positions it for mathematics, hard reasoning, and scientific analysis rather than quick Q&A.

Who it's for: Researchers, mathematicians, and analysts whose workday includes genuinely hard reasoning where a wrong answer costs more than a slow one.

Key features

Agents Multi-agent reasoning

4-8 internal agents tackle the problem in parallel with divergent strategies, then debate and selectively share findings.

Sigma Benchmarks leader

Won or tied first place on Humanity's Last Exam (50.7%) and AIME 2025 math at launch among public reasoning models.

Clock Multi-minute compute

Responses can spend several minutes of extra test-time compute on heavy reasoning when the query warrants it.

X Native X integration

Ships inside SuperGrok subscriptions on grok.com and the Grok iOS/Android apps, with DeepSearch built in.

The honest take

✓ What works

Highest public scores on hard reasoning benchmarks like HLE and AIME among 2025-2026 launches
Multi-agent ensemble reduces single-path failure modes on math/proof work
Larger reliably usable context than Grok 3 across coding and analysis sessions
DeepSearch available inline without a separate tool call

✗ What doesn't

Heavy tier is slow by design — expect multi-minute answers on hard prompts
Most expensive consumer LLM plan at roughly $300/mo SuperGrok Heavy
Rate limits on Heavy reasoning are tight even for paying users
Web browsing is weaker than Gemini 3 Pro Deep Research or Perplexity Max

Verdict

The brute-force compute champion — best when the answer has to be right, not fast or cheap.

💡 Transparency: This review contains affiliate links. If you sign up through our link, we may earn a commission at no cost to you. We only recommend tools we use ourselves. Full disclosure.