Grok 4.20 Multi-Agent Beta
The beta multi-agent endpoint uses parallel agents for research work. The service describes it separately from the single-model Grok 4.20 endpoints.
Preview
What the release provides
Its card lists text and image input, reasoning, function calling, structured output and a one-million-token context. The release is labeled beta.Grok 4.20 Multi-Agent Beta API model card
How to use it well
Practical guidance. Use a multi-agent answer as a draft for review, especially when agents share the same sources. Agreement between several generated accounts is not independent evidence. Check the final citations and retain the original documents used to resolve disagreements.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Status | Preview |
| Access | API |
| Model identifiers | grok-4.20-multi-agent-0309 |
| Context window | 1,000,000 tokens Published API context limit Grok 4.20 Multi-Agent Beta API model card |
| Input price | 1.25 USD per million tokens Standard uncached input; prompt below 200,000 tokens; higher rate applies to all tokens at the threshold Grok 4.20 Multi-Agent Beta API model card |
| Output price | 2.5 USD per million tokens Standard output; prompt below 200,000 tokens; excludes tools and batch discounts Grok 4.20 Multi-Agent Beta API model card |