xAI: Grok 4.20 Multi-Agent API
Grok-4.20-multi-agent-beta-0309 is the first production-grade implementation of xAI's "Society of Mind" architecture. Instead of a single model generating a response, this system orchestrates four specialized sub-agents—The Captain (Orchestrator), The Researcher (X-Firehose specialist), The Logic Engine (Math/Code expert), and The Critic (Fact-checker)—to solve a single prompt through internal debate and verification. Trained on the Colossus 2 cluster with specialized multi-agent reinforcement learning (MARL), it is designed for "High-Stakes" decision making where zero hallucination is required. It natively integrates real-time visual grounding, meaning it can "watch" a live X video stream and discuss it across all four agents simultaneously to reach a consensus.
- Context window: 2,048,000 tokens
- Max output: 1,000,000 tokens
- Input: text, image
- Output: text
- Reasoning: Supported
- Tool calling: Supported
- File input: Supported
- Released: 2026-03-08
- Knowledge cutoff: 2026-02-28
Frequently Asked Questions
What is the context window of Grok 4.20 Multi-Agent?
Grok 4.20 Multi-Agent supports a context window of up to 2,048,000 tokens.
Does Grok 4.20 Multi-Agent support function calling?
Yes. Grok 4.20 Multi-Agent supports tool / function calling.
Does Grok 4.20 Multi-Agent support reasoning?
Yes. Grok 4.20 Multi-Agent is a reasoning-capable model.
What is the knowledge cutoff of Grok 4.20 Multi-Agent?
The knowledge cutoff of Grok 4.20 Multi-Agent is 2026-02-28.