The short answer
Whizi gives you one credit allowance per month and spends it per message. How much a message costs depends only on which model answered it. Starter gets 400 credits, Pro gets 2,000, and Powerhouse gets 8,000.
Roughly half the catalogue costs 1 credit per message. The expensive end is the frontier tier: Claude Opus 5 and GPT-5.6 Sol are 20 credits each. Nothing else about a message changes the price, so a long conversation with a 1 credit model still costs 1 credit per turn.
The full credit scale
There are fourteen rungs across roughly 297 priced models. The distribution is heavily weighted to the cheap end on purpose.
| Credits per message | Models on this rung | Examples |
|---|---|---|
| 1 | 143 | Whizi AI, GPT-5.6 Luna, Gemini 3 Flash, DeepSeek V3.2, Qwen3.7 Plus, the study and writing agents |
| 2 | 51 | GLM-5, Amazon Nova Lite, Arcee Virtuoso Large |
| 3 | 20 | Canadian CPA agent, Amazon Nova Pro, Seed 2.0 Code |
| 4 | 12 | GPT-5.6 Terra, Claude Haiku 4.5, o3-mini |
| 5 | 8 | o3-mini-high, o4-mini-high, Muse Spark |
| 6 | 8 | Grok 4.6, Mistral Large 2407, Mixtral 8x22B |
| 8 | 12 | Gemini 2.5 Pro, Magnum v4 72B |
| 10 | 20 | Claude Sonnet 5, Claude Sonnet 4.6, GPT-5.2, Gemini 3.1 Pro, Kimi K3 |
| 12 | 4 | GPT-5.3 Codex, Sonar, Sonar Reasoning Pro, Nova Premier |
| 15 | 1 | GPT-5.4 |
| 20 | 12 | Claude Opus 5, Claude Opus 4.5 through 4.8, GPT-5.6 Sol, Sonar Pro |
| 25 | 3 | GPT Chat Latest, Sonar Pro Search, Fugu Ultra |
| 50 | 1 | Claude Fable 5 |
| 80 | 2 | Claude Opus 4 and Opus 4.1, the original generation |
The two 80 credit models are the first-generation Opus releases, which remain the most expensive things in the catalogue to run. The current Opus 5 is 20, a quarter of the price of its own ancestor.
What each allowance buys
The useful way to read the allowance is in messages of the model you actually intend to use.
| Plan | Credits | At 1 credit | At 10 credits (Claude Sonnet 5) | At 20 credits (Claude Opus 5) |
|---|---|---|---|---|
| Starter | 400 | 400 messages | not included | not included |
| Pro | 2,000 | 2,000 messages | 200 messages | not included |
| Powerhouse | 8,000 | 8,000 messages | 800 messages | 400 messages |
Nobody spends an allowance on a single model, so the real arithmetic is a mix. A Pro subscriber who sends 40 Claude Sonnet 5 messages (400 credits) still has 1,600 credits left, which is 1,600 messages on the fast tier. In practice the frontier models are the constraint and the 1 credit tier is effectively unmetered.
The allowance resets at the start of each billing month. Credits do not roll over.
What is not charged in credits
Media generation is counted separately, in its own units, and never touches the credit balance.
| Feature | Starter | Pro | Powerhouse |
|---|---|---|---|
| Image generation | not included | 100 per month | 500 per month |
| AI voice | not included | not included | 500 minutes per month |
| AI video generation | not included | not included | 100 per month |
| AI music generation | not included | not included | included |
This is why the credit table above lists no image or video models. They skip the per model credit gate entirely and are tiered by section instead.
File uploads, web search, side-by-side comparison and sharing a conversation are also not charged in credits. Side-by-side does cost two messages rather than one, because two models each answer, and each answer is billed at its own rate.
Making an allowance last
The credit system rewards routing rather than rationing. A few habits that matter more than the rest:
- Draft on the cheap tier, finish on the expensive one. Getting the prompt right takes several turns, and none of them need a frontier model. Draft on a 1 credit model, then switch the same conversation to Claude Sonnet or Opus for the answer you will actually use. See switching models mid-conversation.
- Do not use side-by-side for routine work. It costs two messages, and most tasks show no meaningful gap between models. Save it for decisions and deliverables.
- Watch the 20 credit rung. Claude Opus 5, GPT-5.6 Sol and Sonar Pro are each worth twenty fast-tier messages. They are worth it when the task is genuinely hard, and pure waste on a lookup.
- Check the badge in the model picker. Every row shows its credit cost before you send, so the price is never a surprise after the fact.
The pattern most subscribers land on is a 1 credit default for everything routine and one expensive model reserved for the handful of messages per week that decide something.
- One allowance per month, spent per message, reset at the billing date
- Cost is set by the model alone, not by message length or conversation length
- 143 of the priced models cost 1 credit
- Claude Opus 5 and GPT-5.6 Sol are 20 credits each
- Image, voice, video and music have their own monthly counts
- Side-by-side costs two messages, one per model
- Credits do not roll over to the next month
Frequently asked questions
Does a longer message cost more credits?
No. The credit cost is fixed per message by the model you selected, so a one line question and a ten page document analysis cost the same on the same model. This is deliberately simpler than token billing: you can predict what a conversation will cost before you start it, which you cannot do when the price depends on how long the answer turns out to be.
How many Claude messages do I get on Whizi?
On Pro, 2,000 credits buys 200 Claude Sonnet 5 messages if you spend the whole allowance there, or 500 Claude Haiku 4.5 messages at 4 credits each. On Powerhouse, 8,000 credits buys 400 Claude Opus 5 messages or 800 Sonnet 5 messages. Most people mix, using a 1 credit model for routine turns so the Claude allowance goes further.
Do unused credits roll over?
No, the allowance resets at the start of each billing month and unused credits are not carried forward. If you are consistently finishing the month with a large unspent balance, the cheaper plan is likely the better fit, and if you are running out in the third week the next tier up costs less than the gap in models suggests.
Why do some models cost twenty times more than others?
Because they cost roughly that much more to run. The multiplier tracks provider pricing rather than a product decision, which is why the spread is so wide: the gap between the cheapest and the most expensive model in the wider market is close to three orders of magnitude per answer. Charging a flat rate would mean either overcharging everyone who uses fast models or losing money on everyone who uses frontier ones.
Can I see how many credits I have left?
Yes, the remaining balance is shown in the app, and each row in the model picker displays its credit cost before you send. When the balance runs out, sending pauses until the reset date and the app offers the next plan up rather than silently failing.