SpaceXAI: Grok 4.6

What problem it solves

Grok 4.6 is filed here as a proprietary model from xAI with a large but not maximal listed context. It is for teams considering that API as the hosted brain, especially if they already live in that vendor's stack.

Who it is for

Callers allowed to send prompts to xAI, who will confirm the model id and the retention terms on xAI's site before they cite our card. You want a hosted model and you are comparing it with other hosted flagships.

Who should skip it

Skip it for air-gapped or policy-bound source. There is no open Grok weight hiding behind this row. Skip it if you need a longer window than the listing shows and your prompt actually uses the difference. A million-token peer may fit when this one does not. Measure, do not assume the score compensates. Skip 4.5 or 4.7 rows unless you meant those ids. Adjacent versions are not aliases.

First three steps

1. Read xAI's current model page for 4.6 and note the price and the data terms.
2. Put it next to one other hosted model you can legally call. Reject on license or window first.
3. Run your real prompt. Keep Grok only if the answer beats the alternative by enough to justify the vendor switch, not because the score was 61.

Alternatives on this site

Claude Opus 5, Gemini Flash, and Qwen Max are the hosted notes to set beside it. An open GLM or Kimi is the alternative when the vendor itself is the problem. The comparison tool will narrate the fields. It will not know your contract.

Questions

Is Grok multimodal on this card? Read the category and the modality field on the page you opened. Do not import that fact from a changelog you half remember.

Does a higher Grok version in the directory obsolete 4.6? Only if you have decided to call the higher version. Update the pin explicitly.