Using Grok in Whizi: models, credit costs and plan gates

The short answer

Whizi carries Grok models from xAI inside its 280+ model catalogue, and what you can open is decided by your Whizi plan alone. Exactly one Grok model is reachable on Pro. Every other Grok identifier in the catalogue requires Powerhouse, and Starter includes no Grok model at all.

Plan	Grok access	Credits per month
Free	Any text model, inside a lifetime cap of 7 messages	0
Starter	None	400
Pro	One Grok model, named in the table below	2,000
Powerhouse	Every Grok identifier in the catalogue	8,000

Free accounts are deliberately not model gated: the chat path returns for the free tier before the model gate is reached, so a prospect can try any text model, Grok included, inside a lifetime allowance of 7 messages that never resets.

The Grok rows, with credits and plan

Credits are what you spend from your monthly allowance. The cost per standard answer is provider list pricing, shown so the rungs can be checked against something outside Whizi. The Grok rows in the price index are:

Model	Credits per message	Plan required	Context window	Cost per standard answer
Grok Build 0.1	3	Powerhouse	256K	$0.002
Grok 4.3	4	Powerhouse	1M	$0.0025
Grok 4.5	6	Powerhouse	500K	$0.005
Grok 4.6	6	Pro	500K	$0.005

Two xAI identifiers on the 4 credit rung carry no row in the price index, so there is no context window or per answer price to publish for them. Neither is on the Pro list, so both require Powerhouse.

Model identifier	Credits per message
x-ai grok-4.20	4
x-ai grok-4.20-multi-agent	4

A standard answer in the Whizi cost index is 1,000 input tokens plus 500 output tokens, priced from OpenRouter list rates read on 20 August 2026. It exists to make providers comparable at one fixed shape of turn. It is not what you are billed: the bill is the subscription, and the meter inside it is credits.

Access resolves by identifier, with no separate provider step in the product: the gate checks the Starter set, then the agent set, then the Pro list, and returns Powerhouse for anything else. That fallback is why a Grok release added to the catalogue tomorrow is Powerhouse only until someone deliberately promotes it into the Pro list.

Selecting a model above your tier returns HTTP 403 with the code tier_upgrade_required and the message "Upgrade your plan to use this model."

What each plan buys in Grok messages

One credit is one message on the Whizi house model, and the credit cost of a model is a fixed integer per message. At 6 credits, the top Grok rows spend six times a house model message on every turn, whether the turn is a one line lookup or a full document. The rule behind that is in the credits reference.

Plan	Price	Credits per month	Messages on a 6 credit Grok model	Messages on a 3 credit Grok model
Starter	$15.99/month, or $10.99/mo billed annually at $131.88	400	No Grok model on this tier	No Grok model on this tier
Pro	$29.99/month, or $19.99/mo billed annually at $239.88	2,000	333	Not reachable on Pro
Powerhouse	$49.99/month, or $34.99/mo billed annually at $419.88	8,000	1,333	2,666

Those are whole messages, and the remainder is not spendable: a turn that does not fit in the remaining balance is refused whole rather than part charged. What that looks like, and what to do about it, is in out of credits.

Credits do not roll over. The balance sums only the usage rows carrying the current period key, and for a monthly plan that key is the UTC calendar month, so unspent credits from the last period are unreachable. Running out returns HTTP 429 with the code message_limit_reached.

Weekly billing is offered on mobile only. Credit metering is currently live on web, so the weekly credit figures in code are not what a weekly buyer is metered on today.

Why the Grok rungs sit where they do

Rungs come off a fixed ladder derived from live provider per token pricing, set out in full in the credits reference. What that arithmetic returns for Grok is a 6: the Grok row on the Pro list sits at its cost true rung, so it is sold with no subsidy. At the anchor, one credit is worth $0.00156 of provider spend.

The 6 credit rung carries eight model identifiers today and two of them are Grok. Not every row on the ladder is priced at cost. The search grounded rows are priced above their token rate because a provider charges a flat fee per search that does not scale with tokens: one search is worth 3.2 credits on its own.

For outside scale: the median priced row in the index answers once for $0.00185. The dearest Grok rows sit at $0.005, a little under three times the median row and about one twentieth of the priciest row at $0.105.

A rung is not permanent. The ladder stopped at 20 until 19 August 2026, when a provider price sweep brought in models that cost more than that. Everything through 20 was left unchanged, so no price already on the ladder moved.

Auto does not route to Grok

Auto is a router row in the picker rather than a model. It reads the message and answers on whichever model suits it, and it chooses from six fixed rungs: quick runs on the house model at 1 credit, rewrite and attachment on a Gemini Flash model at 2, longform on a GPT model at 4, code on a Claude Sonnet model at 10, and reasoning on a GPT model at 20. None of the six is a Grok model, so selecting Auto never produces a Grok answer. To use Grok, pick it in the model picker.

Auto also carries no rung of its own. A turn costs whatever the model Auto picked costs, and the clients show a range for your tier instead of a single number: 1 to 1 credits on Starter, 1 to 10 on Pro, and 1 to 20 on Powerhouse. The range is published rather than a single figure because the endpoint that serves it is unauthenticated and does not know who is asking.

Auto also never routes above your plan. It walks back down the ladder until it finds a row your subscription can open, so a Starter account stays on the house model at 1 credit no matter what it asks.

Moving a conversation into or out of Grok is a per message decision, because the charge is per message on the model that answered. A conversation carries the model it was created with, which is why a model never leaves the catalogue once it has been offered.

What Whizi does not record about Grok

There is no per model vision flag. Whizi does not track which models can read an image, so there is no supported list to publish for Grok specifically. In the mobile app an attachment is forwarded to whichever model the turn is using; on the web the composer extracts document text in the browser before sending. The attachment rung in Auto routes to a Gemini Flash model on the reasoning that an image needs vision rather than a large context window.

There is no per model web search flag. Search is a request level toggle with three modes, off, probe and native, applied to the model you selected rather than stored as a property of that model. With search on, a request first carries a cheap probe tool of about 50 tokens, and only a turn where the model actually calls it is re-issued with real search.

A context window is not a per turn budget. A catalogue model with a context window of 93,000 tokens or more gets a flat budget of 40,000 input tokens and 20,000 output tokens per turn, and models with smaller windows get a proportional budget instead. Every Grok row above clears that threshold, so the spread from 256K to 1M makes no difference to a single turn: the widest window and the narrowest one are sent the same amount of text.

The full catalogue by plan is in the model list, and the complete credit scale is in the credits reference.