Short answer
The available evidence does not establish “Grok 4.1 Expert” as a model or API mode. A November 2025 report uses Arena Expert as a leaderboard label and names the entries Grok 4.1 Thinking and Grok 4.1. Separately, xAI’s historical Grok 4.1 announcement describes Thinking and non-reasoning modes. Those labels do not establish that “Expert” means non-reasoning. (BleepingComputer; xAI’s Grok 4.1 announcement)
For API users, there is an important update: xAI’s migration documentation says the Grok 4.1 Fast API slugs grok-4-1-fast-reasoning and grok-4-1-fast-non-reasoning were retired on May 15, 2026. Requests to those slugs are redirected to grok-4.3, with reasoning_effort set to low for the reasoning slug and none for the non-reasoning slug. That documents a migration for the Fast API variants; it does not establish an API mode named “Expert” or map every Grok 4.1 product label to those slugs. (xAI model-retirement guide)
What the comparison evidence does—and doesn’t—show
xAI’s November 17, 2025 announcement reports 1,483 Elo for Grok 4.1 Thinking and 1,465 Elo for Grok 4.1 non-reasoning on the LMArena Text Arena. These are reported leaderboard results, not a controlled API comparison of latency, token use, or performance on your workload. (xAI’s Grok 4.1 announcement)
A separate BleepingComputer article, also dated November 17, 2025, reports 1,510 for Grok 4.1 Thinking and 1,437 for Grok 4.1 on what it calls the Arena Expert leaderboard. The article does not explain how those figures relate to xAI’s 1,483/1,465 Text Arena results. Keep the leaderboard names and figures separate; neither report is a direct Thinking-versus-Expert API test. (BleepingComputer)
A third-party article describes Thinking versus non-thinking as a trade-off: Thinking uses intermediate reasoning tokens and more inference time, while non-thinking aims for faster responses with fewer tokens. It supplies no measured latency or cost comparison and does not equate non-thinking with Expert. Treat this as that article’s description, not a verified API guarantee. (CometAPI)
Production API: what is documented
| API option or label | What the documentation establishes |
|---|---|
grok-4-1-fast-reasoning |
Retired from the xAI API on May 15, 2026; requests to the slug are redirected to grok-4.3 with reasoning_effort: "low". |
grok-4-1-fast-non-reasoning |
Retired on the same date; requests are redirected to grok-4.3 with reasoning_effort: "none". |
| “Expert” | The sources here do not identify this as a current xAI API model name or setting. The Arena Expert wording in the cited report is a leaderboard label. |
grok-4.7 |
A separately documented current API model. xAI documents reasoning_effort settings of low, medium, high (default), and xhigh; reasoning cannot be disabled for this model. |
The first two rows apply specifically to the Grok 4.1 Fast API slugs named in xAI’s migration guide. They should not be assumed to describe every Grok 4.1 app option or historical model-picker label. (xAI model-retirement guide; xAI reasoning documentation)
For a new deployment, xAI’s current model guide shows grok-4.7 in the API model field and documents a reasoning_effort control. Its model catalogue lists $2 per 1 million input tokens and $6 per 1 million output tokens; the release notes specify different rates above 200,000 prompt tokens. Check the live pricing for your account and request size before estimating costs. These are Grok 4.7 rates—not Grok 4.1 or “Expert” rates. (Grok 4.7 developer guide; xAI release notes)
To compare supported options for production, run the same representative prompts and track task quality, time to a usable answer, input/output tokens, and human-review needs. Then apply the documented rates for the exact model and request size. That is a practical evaluation method, not evidence that an Expert option exists or that one setting wins for every workload.
Bottom line
If “Expert” means the Arena Expert leaderboard, the cited report uses it as a benchmark label, not as the name of a Grok mode. If you mean an API option, the evidence here does not establish a Grok 4.1 “Expert” model or setting. For API users, xAI documents the retirement and grok-4.3 redirects for the two named Grok 4.1 Fast slugs; it does not establish that those slugs cover every Grok 4.1 product option. Verify the exact model and reasoning control in the current API docs before deploying.
Sources
xAI's Grok 4.1 rolls out with improved quality and speed for free
Grok 4.1 Released: How It Crushes Other Models — CometAPI
May 15, 2026 Model Retirement — xAI developer documentation
Reasoning — xAI developer documentation