xAI: Grok 4.20 Multi-Agent: the long-context option, reviewed

The quick read

xAI: Grok 4.20 Multi-Agent from x-ai leans on reach: a 2M context window (#1) that targets whole-repo and long-document work. Priced at $1.25 input / $2.50 output per 1M tokens, it trades on how much you can feed it in one call.

Spec sheet at a glance

By the numbers: 2M context window; multimodal (text, image, and file input); 11 exposed API parameters and structured outputs with reasoning support. Use these to judge fit for agentic, long-context, or structured-output workloads.

Pricing & how it compares

Pricing: $1.25 input / $2.50 output per 1M tokens. Its blended $1.88 is above the top-20 median ($0.66). On AI Hippo the composite board rewards cheap long-context models such as Meta: Llama 4 Scout, so a lower rank here means weaker value-per-token, not weaker capability.

Head-to-head

Its closest board neighbor is Meta: Llama 4 Scout (1.3M ctx, $0.20/1M blended). Side by side, xAI: Grok 4.20 Multi-Agent offers 2M ctx at $1.88/1M blended: context is longer and, on blended token price, it is pricier. Pick between them on whichever axis your workload is bound by.

Where it fits

Best-fit workloads: whole-repo & long-document work, screenshot / PDF / image understanding, structured-output pipelines, multi-step reasoning.

Cost in practice & verdict

A representative 100K-input + 20K-output task costs about $0.17 on xAI: Grok 4.20 Multi-Agent. Verdict: a balanced mid-tier option — good when you want capability headroom without top-tier token prices.

Sources

• AI Hippo unified rankings
• Multi-source token pricing
• OpenRouter models API