Grok Free Tier Limits: What Each Tier Gets Per Day (and the Daily Number Nobody Publishes)

There is no published Grok daily message limit for any tier — free, SuperGrok or SuperGrok Heavy. That is not us dodging the question. Under the system xAI describes in its own Grok FAQ (checked 2026-07-30), paid usage is a compute-metered weekly pool shared across Chat, Imagine, Voice, Build and API, and the free tier runs on separate Chat and Voice limits with their own reset schedule. No numeric per-tier allowance appears on any xAI page we could reach. So this guide does the two things that still work: it lays out what each tier verifiably includes and how its clock runs, then shows you how to derive your own effective message limit from the Usage tab in about two minutes. At the end we price the decision you are actually making — wait, upgrade, or call an API directly — using models we ran ourselves.

The short answer, by tier

Free tier: separate Chat and Voice limits, resetting on their own schedule. xAI publishes no message count for them.

SuperGrok: a weekly usage allowance included with the subscription, spendable across five products. Metered in compute, not messages. No numeric allowance published.

SuperGrok Heavy: the higher paid plan. xAI's FAQ says a higher plan carries a larger weekly allowance at a lower cost per action, but it never states what any named plan's pool is. Heavy appears in the FAQ only as a yearly subscription type.

Developer API key: the one layer where exact numbers are published — requests per second and tokens per minute, per model, keyed to cumulative spend.

The tier table: what each one includes

Rows are the four things people mean when they say "my Grok tier", plus the developer key for contrast. The last column is the important one: it says where your real, current number lives, because it is never on a marketing page.

Read the second column down the page and the shape of the problem is obvious. The only Grok tier with a published, checkable number is the one you reach with an API key. Everything on the consumer side is deliberately expressed as a percentage of a compute pool, which is why every "Grok gives you N messages a day" article has to invent N. The deeper mechanics of that pool — what shares it, what the Usage tab fields mean, what credits cost — are in our Grok message limit and shared weekly usage pool guide; this page is about the tiers and the timer.

Why the daily number does not exist any more

A compute-metered pool cannot be expressed as a message count, because different products spend the pool at wildly different rates. The FAQ is explicit that a chat message uses little compute while generating a high-quality video or running a long coding task uses far more. One pool, five surfaces — API, Build, Chat, Imagine and Voice — and you may spend the whole thing on one of them.

So a daily message number is not a figure xAI is withholding. It is a figure that does not exist under the current system, and it changes meaning depending on what you did on Tuesday. Three consequences worth internalising:

• Your effective message limit is personal. Two SuperGrok accounts on the same plan get different message counts out of the same allowance, because one of them renders video.
• The clock is weekly, not daily. Anything phrased as a per-day cap on the paid tiers is describing the previous system. The weekly-pool rollout began June 2026.
• Third-party counts are stale. Figures like ~1,000 text messages a day for SuperGrok or ~10,000 for Heavy circulate widely, are not confirmed by any xAI page, and predate the weekly pool. Useful as folklore, wrong as a spec.

Which leaves one honest question: what is my number? You can work it out.

Derive your own limit from Settings → Usage

The Usage tab gives you a percentage, and a percentage plus arithmetic gives you a message count. This is the durable version of the answer — it survives every xAI pricing change, because it reads your own account rather than a published table.

The two-minute derivation. Do this once and you will never need a published number again:

1. Open Settings → Usage and write down the percentage. Call it used.
2. Note how many days into your week you are. Call it d. Your burn rate is used / d points per day.
3. Runway in days is (100 − used) / (used / d). Compare that against the days remaining until the reset time shown on the same tab.
4. To make the pool last the full week, your ceiling is 14.3 points per day — that is 100 divided by 7, and it is the only universal number on this page.
5. For a message count: note the percentage, send a measured batch of your normal work — say 20 chat turns — and read the percentage again. If it moved by Δ points, your remaining allowance is roughly (100 − used − Δ) × 20 / Δ turns of that kind — the batch you just sent comes off the top before you divide.

Step 5 is the one nobody does, and it is the only way to get a real message number for your own account. It is also model-dependent and product-dependent: measure it again if you switch between quick chat and long agentic runs, because the compute cost per turn is not the same thing.

How long until your usage replenishes

The paid pool resets once a week at a date and time specific to your account, shown in Settings → Usage. xAI publishes no global reset day, hour or timezone. That tab is the timer. There is no other authoritative source for it, and any page quoting you a fixed reset hour for everyone is guessing.

Four clocks run at once, and confusing them is why people think their timer is broken:

• The weekly pool clock. Weekly, account-specific, shown as a date and time. What the anchor is tied to — signup, billing date, first use — is not documented, so treat the tab as ground truth rather than reverse-engineering it.
• The free-tier Chat and Voice clock. Separate limits on a separate schedule, per the FAQ. Hitting the weekly pool does not consume these, and they replenish independently. This is why a paid user who runs dry still gets a working chat box.
• The credits clock, which does not tick. Extra Usage Credits are not refilled by the weekly reset. They sit until spent and expire one year after purchase.
• The API clock, which is not weekly at all. Developer keys are governed by rolling per-second and per-minute windows. If your code got an HTTP 429, you did not exhaust a weekly pool and no reset date will help you — see our rate-limit fix guide.

If the number you want is "hours until I can send again", subtract now from the reset time on the Usage tab. That is the whole method, and it is exact for your account in a way no published table can be.

What to do while you wait

Hitting the weekly limit pauses paid features until the reset. It is a downgrade, not a lockout — your account stays active, your history stays put, and the free-tier Chat and Voice limits remain available on their own schedule. From there you have five moves, roughly in order of what they cost you:

1. Wait. Free. The reset date and time is on the Usage tab. If your runway maths says you are three days short, this is the option you should at least price against the others.
2. Use the free tier in the meantime. Per the FAQ you keep free-tier Chat and Voice access. For light questions this is often enough to bridge a couple of days.
3. Buy Extra Usage Credits. $5 minimum, web only, spent immediately since the pool is already empty. The catch is stated by xAI itself: credits bill at standard rates, which is a higher cost per action than the effective rate your subscription gives you. Topping up repeatedly is the most expensive way to buy Grok compute.
4. Upgrade the plan. The FAQ says a higher plan means significantly more weekly usage at a lower cost per action than repeatedly buying credits. It does not say how much more, so you are buying an unquantified allowance — which is fine if your derivation says you are 20% short, and a gamble if you are 200% short.
5. Move the work to an API. Only sensible if the work is programmatic rather than conversational — you cannot route a SuperGrok chat elsewhere. But if what you were doing behind that limit was generating code, this is the option with an actual price tag, and it is the next section.

For the app-side workflow questions this raises, how to use Grok covers the product surfaces, and free LLM APIs covers what a zero-cost bridge actually looks like.

Upgrade, or call an API: the costed version

If the work behind your limit is code generation, "wait, upgrade, or route elsewhere" stops being a shrug and becomes arithmetic. This is the one part of this page that is ours rather than xAI's. We ran 23 models through a single harness of nine executed Python tasks — the original sweep was 13 models run in one sitting, the other 10 ran later on the same harness with the same settings — and 20 of the 23 scored 9/9.

These are coding-task measurements, not chat measurements. Nothing here tells you how Grok answers a question in the app. It tells you what a bounded code-generation task costs on an API, which is the alternative you are weighing when a subscription runs dry mid-project.

Three readings of that table, in the order they matter to someone staring at an empty pool:

• Grok 4.5 scored 9/9 at a measured $2.93 per 1,000 tasks. It is a model that does the work. If you want Grok specifically, the API is the version of Grok with a published price and a published limit — rates in our Grok API pricing guide, and the full result in the Grok 4.5 review.
• Claude Haiku 4.5 matched that 9/9 at $0.94 — 3.1x cheaper and faster, at 3.7 s against 6.6 s. GPT-5.4 also went 9/9, at $1.69 and 3.6 s. On nine bounded Python tasks these three are indistinguishable on correctness, so the money is the only axis left.
• Grok 4.3 is the cheaper Grok and it missed one. 8/9 at $1.75, dropping flatten. Details in the Grok 4.3 review. Cheaper is not automatically the downgrade you want when a $0.94 model scored a point higher.

The spread from the cheapest 9/9 in that table to Grok 4.5 is 29x — $0.10 against $2.93 per 1,000 tasks. So the honest decision rule: if the work is bounded code generation, an API key costs cents and has a published ceiling, which is a strictly better position than an unquantified weekly pool. If the work is conversation, image or voice, none of this applies and you are choosing between waiting and paying. More cheap options in best cheap LLM for coding, the whole field ranked four ways in the AI coding ranking, and the Haiku result in full in the Claude Haiku 4.5 review.

How our numbers were produced

Nine executed Python tasks: two_sum, valid_parentheses, merge_intervals, roman_to_int, lcs_len, flatten, top_k_words, token_bucket, parse_csv_line.

The model receives a function signature and a prose spec. It never sees the assertions. The returned code runs against hidden assertions in an isolated python3 -I subprocess with a 12-second timeout. Every assertion passes or the task fails — no partial credit, no human grader, no LLM judge. Temperature 0, max_tokens 4000, one scored attempt per task, retried only on an API error.

Cost is derived, not invoiced. It is the token counts the API reported multiplied by that model's list price on a stated date. Prices move — 49 of roughly 396 catalogue models repriced in the twelve days to 2026-07-29 — so an undated cost figure is not a fact. Reasoning-token counts in the table are per task, not totals.

Calls go through OpenRouter's OpenAI-compatible endpoint, deliberately not through the DataLLM Lab gateway. Nothing on this page depends on our infrastructure. The core sweep was 13 models run in one sitting; the remaining 10 ran later on the same harness under the same settings, for 23 in total.

What we did not measure, and what we could not source

Not measured at all: long-context reasoning, multi-file refactoring, agentic or multi-turn tool use, any non-Python language, vision, translation quality, and anything run locally. The harness is single-turn and API-hosted. A 4,000-token ceiling can truncate a very verbose model, and a truncated answer scores as a miss.

Not tested, and never claimed as ours: Grok 3, Grok 4, Grok 4.20, grok-build-0.1, Claude Fable 5, GPT-5 nano, VibeThinker, any Ollama or locally-run deployment, and any Gemini besides 3.6 Flash and 3.1 Pro. Grok 4, Grok 3, Grok 3 Mini and Grok Code Fast 1 are retired and should never be priced as callable models.

Not sourceable, and therefore not stated: any per-tier daily or weekly message count for free, SuperGrok or SuperGrok Heavy; the subscription price of each tier; the per-action or per-token rate of an Extra Usage Credit; and the anchor that sets your weekly reset moment. We checked xAI's Grok FAQ and rate-limits pages on 2026-07-30 and none of those appear there. If you find a page quoting them, check whether it cites xAI or another blog.

And one thing that will rot: consumer limits change. The tier table above is a snapshot of what xAI documented on 2026-07-30. The derivation in Settings → Usage is the part written to outlive it.

FAQ