Grok models explained

xAI positions Grok 4.5 as its most capable and fastest model and recommends it as the default for most work, including code. It has a 500k context window. Prices shown are the standard tier for prompts under 200k tokens; larger prompts are billed at a higher rate.

Grok 4.3 is no longer the flagship but remains available at a lower price than 4.5, with a larger 1M context window. It suits high-volume or very long-context work where the newest model is not required.

The 4.20 reasoning family suits harder prompts where careful intermediate work matters. Verify the exact suffix in the official docs before using it in production.

The non-reasoning 4.20 variant appears in the API pricing table. Treat it as a versioned API option and confirm access in the xAI console before building around it.

Use multi-agent style models for occasional difficult tasks, not routine high-volume traffic. More complex runs can still cost more through longer outputs.

Grok Build is positioned for agentic coding and development tasks. It has a smaller context window than Grok 4.3 but lower listed output pricing.