How to use Grok 4.7 on the web: Expert, Fast, and upgrade prompts

Start with the current mode menu and identify what you opened

Open grok.com, sign in, and find the button showing the current mode. The public page checked for this article displayed Fast; the button opens the mode menu. Expand it to see the options available to your account. If Expert is available, select it and verify the current mode label. This article reviews public pages and official material; it has not signed in or tested a paid account on readers' behalf.

Record three things together: the URL, signed-in identity, and current mode. Menus in someone else's screenshot, button positions in older tutorials, and your actual interface may differ. If you only see a mode name, first establish whether it appears in the normal chat input area or a developer tool's settings page, so you do not investigate the wrong product from the outset.

Start with a short question containing no private information and see whether the request completes normally. Do not use “What model are you?” as the sole acceptance check: the answer is generated conversation text and cannot replace the account interface and product announcement. Reply length and waiting time cannot establish the backend version either.

Let the interface establish what a free account can access

Record the menu's available options, unavailable options, and prompts separately before deciding whether to pay. Clarify any conditions you do not understand; there is no need to purchase immediately just to reproduce someone else's screenshot.

For example, a friend can select a mode while your page shows Upgrade. That comparison still lacks both users' signed-in identities, subscriptions, and clients; it does not establish that your account is broken. Filling in those details helps locate the difference more than repeatedly refreshing or switching accounts.

Before paying, write down your needs: do you occasionally organize a short text, or check complex plans every day? Complete a representative task through your currently available entry point and list concrete gaps. If the gap is merely output formatting or incomplete source material, improve the input first. Do not treat every unsatisfactory answer as proof that you must upgrade.

Paid but still asked to upgrade: compare purchase and current login identities

The FAQ says a subscription is tied to the purchasing account. For web purchases, check Settings → Billing. For Apple or Google Play purchases, check the relevant store first, then verify the current signed-in identity.

Record the payment channel, purchasing account, and account currently in use separately. If the payment belongs to another login identity, sign in through the original method before buying again. Keep only the parts of screenshots needed to explain the problem; redact full email addresses, order numbers, and other personal information, and retain the original receipt privately.

Next, record when the upgrade prompt appears: when opening the page, expanding the menu, or after submitting a message. Different stopping points call for different evidence. Resolve identity mismatches before investigating subsequent requests. Do not change accounts, devices, and subscriptions simultaneously, because that makes it difficult to tell which change affected the result.

A visible mode that cannot continue: distinguish usage limits from task failures

Check the percentage used and reset time in Settings → Usage. The FAQ says paid features pause until the reset once the weekly paid allowance is exhausted.

Record the original failure text in two categories: prompts explicitly mentioning usage, resets, or subscriptions, and failures to submit messages, page errors, or answers that do not meet requirements. Check account status first for the former. For the latter, retain the error and time and retest with short text. A calculation mistake in an answer and an account unable to send requests are different problems.

For example, if short text completes but a long document consistently fails, reduce the input in a copy without sensitive information, then add content gradually. This is a troubleshooting suggestion, not a claim that any fixed file size or length must be supported. Change one condition at a time to identify whether content changes or entry-point status caused the difference.

Compare the same small example without treating one timing result as a conclusion

Prepare a small example of your own: option A has a CNY 80 service fee and CNY 20 shipping; option B has a CNY 95 service fee and free shipping. Ask for a total-price comparison using only those figures, without adding tax or discounts, followed by two detail rows and one concluding sentence. The expected totals are CNY 100 for A and CNY 95 for B, with B cheaper by CNY 5.

Use the same input in each available mode and check three things: whether it adds assumptions, calculates correctly, and follows the specified format. For a real business comparison, prepare independently verifiable material and keep the correct answer or acceptance criteria outside the conversation.

A comparison log needs only four columns: input task, current mode, actual result, and required corrections. For example, correct arithmetic without the requested two-row format is a formatting issue; inventing a nonexistent discount is an extra assumption. Adjust only the corresponding requirement next time, rather than changing both the material and acceptance criteria. This gives you a basis for choosing a mode for your work, without presenting one demonstration as a general performance conclusion.

One result describes only that particular task. To assess ongoing use, try several kinds of actual work samples and record rework and manual corrections; do not retain only the most polished reply. This is a comparison method, not a tested ranking of the two modes or a speed guarantee.

Mobile, Bot, and developer tools: identical names do not establish identical eligibility

The official overview says web, iOS, and Android can synchronize conversations, settings, and subscriptions. That does not establish that every new feature appears simultaneously. If the mobile interface differs, first check the web entry point confirmed here, then consult the relevant client documentation.

The same update also lets SuperGrok Plus and Heavy users open threads through the Bots section in the web sidebar. That eligibility does not describe every Bot client. Do not use the Bots menu to determine whether the correct mode is selected for normal chat.

The Grok 4.7 Fast variant in developer documentation is listed only in Cursor and Grok Build, billed under those plans, and unavailable through the public xAI API. Its token rates cannot be applied to web Fast. Scripts, coding tools, and web chat each need their own billing and documentation checks.

If the problem remains unexplained, seek help with reproducible steps

For web issues, the FAQ suggests trying an incognito window in Chrome or Edge, or seeking help through Report an issue in the product.

A useful record states the time, client, URL, selected option, and resulting prompt. Add the shortest reproducible task without sensitive information. Describe “I cannot see the mode” separately from “the mode is visible but the request fails,” so support does not have to guess where progress stops.

For example: “I opened the website in a browser, and the mode button showed this name. Clicking it produced this prompt. The same short text completes through another available entry point.” Replace the names and prompts with your actual observations and mark unverified steps. If you have no comparison result, say it is unverified instead of guessing to make the report appear complete.

If there is no immediate conclusion, keep using the tasks that currently work and wait for a clear response before changing your subscription or workflow. Do not upload a complete set of business material to prove a model upgrade, or pay repeatedly to test it. This article helps identify the entry point and failure stage; current product prompts and official support must establish actual account eligibility.