Grok 4 Heavy Paid

Grok 4 Heavy is xAI’s multi-agent mode: four Grok 4 instances work a problem in parallel, compare notes, then produce a single answer. It launched at $300 per month as the top tier of SuperGrok Heavy, and it targets hard research, math and long-form analysis rather than everyday chat.

What is Grok 4 Heavy?

xAI first demoed the multi-agent setup when Grok 4 launched in July 2025, showing four agents coordinating on the Humanity’s Last Exam benchmark. Heavy is the paid product version of that idea. Instead of one model thinking longer, you get several reasoning agents that can cross-check each other’s chains, which reduces the chance of a confident wrong answer on multi-step problems.

How does Grok 4 Heavy actually work?

• Four Grok 4 agents start from the same prompt and work independently in parallel
• An aggregator compares their reasoning traces and resolves disagreements
• Native tool use: web search, code execution and X data access
• 256K token context window across the agent pool
• Available through grok.com, the X app and the xAI API
• Billed at $300/month on the Heavy plan, or per token via API

What is it good at?

Multi-agent helps most when a task has a verifiable answer that is easy to get subtly wrong. Competition math, financial modelling with many assumptions, and literature reviews all fit. On simpler asks the extra agents add latency without adding much, and xAI’s own comparisons show the benchmark gap between Grok 4 and Heavy is real but not enormous.

• Self-consistency: wrong reasoning in one agent tends to get flagged by the others.
• Deep search: it can run several search strategies at once instead of one query chain.
• Code plus math: it writes and runs Python mid-reasoning to check results.

Who actually needs it?

1. Quantitative analysts pressure-testing a model with many edge cases.
2. PhD students benchmarking against frontier reasoning tasks.
3. Due-diligence teams reading dozens of filings at once.
4. Engineers debugging a rare race condition with a long reasoning trail.
5. Anyone who needs a second opinion baked into the same answer.

Is Grok 4 Heavy worth $300 a month?

For a single power user, no — standard Grok 4 or a competing reasoning model covers most days. For a small research team that would otherwise burn analyst hours, the math can flip. Two real drawbacks: it is slow, since four agents take longer than one, and you cannot inspect the individual agent transcripts in the consumer app, so the ensemble stays something of a black box.

Quick Summary

• Grok 4 Heavy runs four Grok 4 agents in parallel and merges their reasoning.
• It launched at $300/month via SuperGrok Heavy and is also on the xAI API.
• It excels at verifiable multi-step tasks like math, code and deep research.
• It is slower and pricier than single-model Grok 4, and agent traces stay hidden in the app.
• Best for research teams, not casual users.