xAI releases Grok 4.20 with a multi-agent mode

Background

The Grok 4.20 release came at a time of change for the company: xAI became a SpaceX subsidiary in the same month. The model itself was a step within the Grok line, and it introduced a multi-agent option.^[1]

What happened

Grok 4.20 was released in February 2026.^[1] xAI's release notes record Grok 4.20 and Grok 4.20 Multi-agent going live in the API in March 2026.^[2] The multi-agent model has several agents collaborate in parallel on deep-research tasks.^[3]

Key details

The model has a 1-million-token context window and takes text and images as input.^[3] The API price was $1.25 per million input tokens and $2.50 per million output tokens for prompts under 200,000 tokens.^[3]

The company also released Grok 4.3 in April 2026.^[1]

Several terms explain the release. A context window is the amount of text a model can keep in view, measured in tokens. A multi-agent system is one in which several model instances each take part of a task, and a coordinating step combines their results. Deep research is a long task of searching, reading and synthesising sources into a report. Multi-agent designs can cover more ground in the same time, but they usually use more computing per request, so their price and speed are worth checking before use.

Pricing is charged per million tokens, so a long research task that reads many sources will cost far more than a short question. When judging a release like this, it helps to separate what the company announced, what developers can actually use, and what independent testing has shown.

Impact

The release made a multi-agent option available through the API, so developers could use parallel agents for research-style work rather than building that arrangement themselves. For readers, the event matters because it marks a change from single-model answers to models that coordinate several agents on one task.