Using GPT in Whizi: models, credit costs and plans

The short answer

GPT models are part of the Whizi catalogue and are reached with a Whizi subscription, not a ChatGPT one. The cheapest GPT row costs 1 credit per message and is included on Starter. The rest sit on Pro or Powerhouse, and the most expensive GPT row in the catalogue costs 25 credits per message.

The default assistant, Whizi AI, is a house persona that runs on the base GPT model, so every paid plan reaches GPT from the first message. One credit is defined as one Whizi AI message, which means one credit is also one message on that base GPT model.

Plan	Price	Monthly credits	GPT models reachable
Free	no subscription	0 credits, 7 message lifetime allowance	any text model in the catalogue, inside that allowance
Starter	$15.99/month, or $10.99/mo billed annually at $131.88	400	GPT-5.6 Luna, plus Whizi AI which runs on it
Pro	$29.99/month, or $19.99/mo billed annually at $239.88	2,000	GPT-5.6 Terra, GPT-5.5, GPT-5.2, GPT-5 Mini, GPT-4.1, GPT-4.1 Mini, GPT-4o, GPT-4o Mini, GPT-OSS-120B
Powerhouse	$49.99/month, or $34.99/mo billed annually at $419.88	8,000	everything above, plus every other GPT row in the catalogue

What each GPT model costs

A credit cost is a fixed integer per model, and the mechanics of the scale are in how credits work. The OpenAI rows in the catalogue run from 1 credit to 25, so the cheapest GPT answer and the dearest differ by a factor of twenty five on the same bill.

The dollar column is what one answer costs at provider list rates, measured on a standard answer of 1,000 input tokens plus 500 output tokens, using rates read on 2026-08-20. It is there so the credit column can be checked against something real, not because you are billed in dollars per message.

Model	Context window	Credits per message	One standard answer	Plan
GPT-5.6 Luna	1M	1	$0.0008	Starter
GPT-5.4 Nano	400K	2	$0.000825	Powerhouse
GPT-5.4 Mini	400K	4	$0.003	Powerhouse
GPT-5.6 Terra	1M	4	$0.008	Pro
GPT-5.4	1M	15	$0.01	Powerhouse
GPT-5.6 Sol	1M	20	$0.01	Powerhouse
GPT-5.5	1M	20	$0.02	Pro
GPT Chat Latest	400K	25	$0.02	Powerhouse

Further GPT rows carry a credit cost but are not in that price set:

Model	Credits per message	Plan
GPT-4o	1	Pro
GPT-4.1 Mini	2	Pro
GPT-5 Mini	2	Pro
GPT-4.1	8	Pro
GPT-5	8	Powerhouse
GPT-5.1	8	Powerhouse
GPT-5.1 Codex	8	Powerhouse
GPT-5.1 Codex Max	8	Powerhouse
GPT-5.2	10	Pro
GPT-5.2 Chat	10	Powerhouse
GPT-5.2 Codex	10	Powerhouse
GPT-5.3 Codex	12	Powerhouse

The OpenAI reasoning series sits on Powerhouse:

Model	Credits per message	Plan
o3-mini	4	Powerhouse
o4-mini	4	Powerhouse
o3-mini-high	5	Powerhouse
o4-mini-high	5	Powerhouse
o3	8	Powerhouse

Three OpenAI rows are charged below what the cost formula returns, by owner policy rather than arithmetic. The rung scale itself, and the rule for a model that lands between two rungs, are in how credits work.

Model	Charged	What the formula gives
GPT-5.6 Terra	4 credits	10
GPT-5.5	20 credits	25
GPT-5.6 Sol	20 credits	25

Which plan unlocks which GPT model

Access is resolved on the server by model identifier. On Starter the gate falls on every GPT row above the base 1 credit model. On Pro it falls on the OpenAI identifiers outside Pro's nine, which is where the Codex coding rows, the o-series and the top GPT row sit. In both cases the refusal is HTTP 403 with the code tier_upgrade_required and the message "Upgrade your plan to use this model."

Starter reaches exactly four picker entries: Auto, Whizi AI, the base GPT model, and one fast Gemini model. That is the whole list, so on Starter every GPT message costs 1 credit and nothing above the base row is available.

Pro adds nine OpenAI entries, the ones named in the short answer table above. The set pairs each included flagship with its mini variant, and adds the open weight GPT-OSS row. Pro stops there: the Powerhouse trophy GPT row, the Codex coding rows and the o-series are not in the set. Pro is a strict superset of Starter, so the Starter GPT row is included as well.

Powerhouse is the fallback tier, not Pro. Any model identifier that is not in the Starter set and not in the Pro list resolves to Powerhouse, which is why the coding specialist GPT rows and the o-series sit there. It also means a GPT row the catalogue gains arrives on Powerhouse and stays there until it is deliberately promoted, so the nine names in the Pro column are the whole of what Pro reaches on the OpenAI side, not a sample of it.

The dearest GPT row is held back on Powerhouse deliberately rather than by the fallback rule: it is the trophy the top tier sells. The picker shows the consequence. Its Recommended section is limited to models a Pro subscriber can already reach, so the OpenAI slot there is the Pro tier GPT row instead.

What an allowance buys in GPT messages

A GPT turn either fits in the remaining balance or it is refused outright: a 25 credit GPT turn on a balance of 20 credits is rejected, not discounted down to what is left.

Plan	Monthly credits	At 1 credit	At 4 credits	At 20 credits
Starter	400	400 GPT messages	not included	not included
Pro	2,000	2,000	500	100
Powerhouse	8,000	8,000	2,000	400

Weekly billing, offered on mobile only, carries its own numbers: Starter 100 credits, Pro 500, Powerhouse 2,000 per week.

Every model a Starter account can reach costs 1 credit, so on that tier 400 credits is 400 messages.

Credit pricing is enabled per platform. The web app is on the allowlist today and is charged the real multipliers against the credit allowance. A client that is not on the allowlist is charged one per turn against the older message allowance instead, which is 400 monthly on Starter, 800 on Pro and 5,000 on Powerhouse. A client outside the allowlist also draws no credit badges in the picker, because the cost field is absent rather than pinned to 1.

Running out returns HTTP 429 with the code message_limit_reached and the message "Your monthly message limit has been reached." Weekly plans say weekly. What to do about it is in running out of credits.

Using GPT without a ChatGPT subscription

A free Whizi account gets a lifetime allowance of 7 messages that never resets, and the free tier is deliberately not model gated. Any text model in the catalogue can be tried inside that allowance, GPT included.

When the free allowance is spent, the response is HTTP 402 with the code free_limit_reached and the message "Your free messages are used up. Start a subscription to keep chatting."

After that, a paid plan is what unlocks the ongoing allowance. New accounts start with a 7 day trial charged at the intro rate of $0.99, after which the selected plan renews unless it is canceled.

Existing ChatGPT history can come with you. Whizi imports ChatGPT and Claude exports, and the export file itself never reaches the server: the browser parses it and posts normalized conversations. One request carries up to 25 conversations and 400 messages per conversation, and anything larger is split into follow up slices automatically. A raw .json export is accepted up to 300 MB, past which the .zip export is the route. Imports are idempotent, so clicking Import twice or resuming after a dropped connection lands the same chats once.

Two things do not survive a ChatGPT import: only the branch that was on screen is imported, so regenerated replies and discarded drafts are not replayed, and non transcript content is skipped, which covers hidden reasoning, analysis tool code and output, browsing scratchpads, custom instruction blocks and function calls. If an import stalls or reports skipped conversations, see chat import errors.

Moving between GPT and the other families

A credit cost is looked up per model per turn, so the price of the next answer is set by whichever model answers it. Moving a conversation from GPT to another family changes what the next turn costs and leaves the turns already sent alone. A conversation also carries the model it was created with, which is why a model is never removed from the catalogue once it has been offered.

Auto is the other way to reach GPT. It is a router row, not a model, and it resolves to a concrete model before anything is charged, so the credit charge, the tier gate and the stored history all record the model that actually ran. Its ladder is six rungs. The cheapest lands on the house model at 1 credit, and two land on OpenAI rows: long form at 4 credits and multi step reasoning at 20 credits.

Auto never routes above your plan. That behavior, and the other reasons a row you expect is missing from the picker or refuses to send, are in model unavailable.

What a GPT model in Whizi does not do

The context window column above is the provider window, not your per turn budget. Most catalogue models get a flat budget of 40,000 input tokens and 20,000 output tokens per turn, and a model whose window sits below 93,000 tokens gets a smaller proportional budget instead, with output capped at 40 percent of the window and a 1,000 token safety margin held back. Every GPT window in the first table above is 400K or 1M, far above that threshold, so those rows all take the flat budget: a million token window and a 400K one hand you the same 40,000 tokens per turn. A conversation that outgrows the budget is covered in context too long.

Web search is not a GPT capability flag. It is a per request toggle with three modes, applied to whichever model the turn is on, so there is no list of GPT models that can or cannot browse. With the toggle on, the request first carries a cheap probe of about 50 tokens, and only a turn where the model actually calls it is re-issued with real search. Native provider search injects a fixed preamble of roughly 4,400 input tokens, which is why the probe exists.

Sending speed is not a GPT setting either. Chat is capped at 10 messages per minute and 60 per hour per user, and those two ceilings are identical on every tier, so no GPT row and no upgrade moves them. The full set of windows is in rate limits.