Grok 4.20 ships multi-agent, 2M context, weekly updates

xAI released Grok 4.20 in public beta on February 17, with a second iteration on March 3 [xAI news].

── What shipped ──

• Multi-agent orchestration. Tasks can be delegated to sub-agents that work in parallel.
• 2M-token context window. Largest among the major closed-weight vendors.
• Weekly-cadence updates. xAI is shipping incremental capability bumps faster than any other major lab.
• Hallucination rates down to 4.2%, per xAI's own benchmarking [AI Business].

The model is available to SuperGrok and Premium+ subscribers and via the xAI API.

── Why it matters ──

The multi-agent capability is the most consequential feature. Almost every frontier vendor is racing to ship native agent orchestration; doing it inside the model rather than as a wrapper is a different architectural bet.

The weekly-update cadence is the more honest signal. xAI has been the noisiest lab on social, but the actual shipping rhythm has now caught up to the talk. For developers, this means features land fast — and break more often. Production deployments need to pin model versions explicitly.

The hallucination claim is unverified by independent benchmarking. Treat as a marketing number until corroborated.

── Editor's take ──

Grok continues to be the frontier model with the highest social-noise-to-substance ratio, but 4.20 is the first version where the substance is non-trivial. If you've dismissed xAI on brand, the multi-agent and 2M context features alone now warrant a re-evaluation. The corporate situation — xAI absorbed into SpaceX, IPO machinery active — is its own story.