The short answer
Yes, Whizi carries Meta's Llama family inside its 280+ model catalogue, and both priced Llama rows sit on the cheapest rung Whizi meters: 1 credit a message. Llama sits on the Pro plan, which is $29.99/month, or $19.99/mo billed annually at $239.88. Starter does not include any Llama model.
The Llama rows Whizi prices, with the credit cost charged per message:
| Model | Context window | Cost per standard answer | Credits per message | Minimum plan |
|---|---|---|---|---|
| Llama 4 Maverick | 1M | $0.0006 | 1 | Pro |
| Llama 3.3 70B Instruct | 131K | $0.00026 | 1 | Pro |
| Llama 4 Scout | Not carried in the Cost Index | Not carried in the Cost Index | Not published | Pro |
A standard answer is 1,000 input tokens plus 500 output tokens, priced from rates read on 2026-08-20, so the figures compare like for like across providers.
Those three identifiers are the Llama rows named in the Pro model set. Any other Llama identifier the catalogue carries falls to the Powerhouse fallback described below, so treat the table as the Pro list rather than a complete catalogue enumeration.
What a Llama message costs in credits
Both priced Llama rows sit on the bottom rung of the credit ladder, so a Llama turn costs 1 credit whatever you send it. The third Llama identifier in the Pro set is not carried in the Cost Index, so this page publishes no credit figure for it.
At 1 credit a message, the credit allowance is the Llama message count: on the web, Pro buys 2,000 Llama turns a month and Powerhouse buys 8,000. Starter reaches no Llama row at all.
| Plan | Credits per month (web) | Messages per month off the credit allowlist |
|---|---|---|
| Starter | 400 | 400 |
| Pro | 2,000 | 800 |
| Powerhouse | 8,000 | 5,000 |
Weekly billing is offered on mobile only, and not on every tier. Where it is sold, the weekly credit allowances are 100 on Starter and 500 on Pro, and the weekly message allowances are 100 and 400.
A turn is charged at the rung of the model that answered it, so a conversation that moves between families is billed per turn at each answering model rate. Credits do not roll over: the balance sums usage inside the current period key, so the arrival of a new period is itself the reset. There is no top up path.
Credit multipliers are charged only on platforms in the credit allowlist, which today is the web app. A platform outside the allowlist sees a flat 1x, is charged one per turn, and keeps the older message allowance instead. On Starter the two allowances are the same number, 400. On Pro and Powerhouse they are not, which is the second and third columns of the table above.
The full ladder, all fourteen occupied rungs, is in the credits reference.
The plan gate, exactly
Access is resolved server side by identifier (see the models reference).
Starter reaches exactly four picker entries: Auto, Whizi AI, one OpenAI model and one Google model. No Llama row is among them.
Pro adds the Llama set on top of Starter. Read as the curation rule applied to one family, that set is the current Llama 4 rows plus the older Llama 3 workhorse, three identifiers in total.
Powerhouse includes everything Pro reaches, because the tier ranks compare with a greater than or equal test, so a higher tier always satisfies a lower requirement. Powerhouse is also where any catalogue identifier lands that is not named in the Starter or Pro sets, which is why an unlisted Llama identifier is Powerhouse until it is deliberately promoted.
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." That symptom is covered in when a model is unavailable.
Free accounts are not model gated. The chat path returns for the free tier before the model gate is reached, so a free account can try any text model, limited by a lifetime allowance of 7 messages that never resets. The free credit allowance itself is 0, so those 7 messages are the whole of it.
Context window versus what a turn actually gets
The context window in the table above is the model published window. It is not the budget a single Whizi turn runs with.
Every catalogue model gets a flat budget of 40,000 input tokens and 20,000 output tokens per turn. 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.
Because both priced Llama rows sit above the 93,000 token threshold, both run on the same flat 40,000 input and 20,000 output budget. The larger published window does not raise what a single Whizi turn sends, and the provider counts requested maximum output against the window as well as input, so the 1M row and the 131K row are sized identically here. If a thread outgrows the budget, see when a conversation is too long.
Where these rows sit in the Cost Index
The Cost Index prices 100 rows across 29 providers at one standard answer each. That number is the size of the price index, not the size of the Whizi catalogue.
| Reference point | Cost per standard answer |
|---|---|
| Cheapest priced row in the index | $0.000053 |
| Median priced row | $0.00185 |
| Llama 3.3 70B Instruct | $0.00026 |
| Llama 4 Maverick | $0.0006 |
| Priciest priced row in the index | $0.105 |
Both priced Llama rows land below the index median, and the spread between the cheapest and priciest priced row is roughly 2000x. 89 of the 100 priced rows carry a Whizi credit charge; the rest are priced for comparison but are not offered as their own metered row.
What Llama is not used for in Whizi
No Llama row sits on the Auto ladder. Auto has six rungs, and the models on them are the house model for short questions, a Gemini Flash row for rewrites and again for attachments, a GPT row for long form, a Claude Sonnet row for code and a second GPT row for multi step reasoning. No Llama identifier appears on any of the six, so Auto cannot reach Llama at any tier and a Llama answer is always a manual pick in the model picker.
Attachments on Auto go elsewhere. Auto routes any turn carrying an attachment to Gemini Flash, on the reasoning that an image needs vision rather than a large window. Select a Llama row yourself and the attachment is handled per request rather than per model. On the app, it is forwarded to whichever model the turn is on, and only the most recent user message carrying attachments is forwarded. On the web, a PDF, Word or spreadsheet file has its text extracted in the browser and the text is what reaches the model.
Voice and media generation are not per model choices. Media generators skip the per model gate entirely and are tiered by picker section instead, and voice mode runs on its own realtime route, so picking a Llama row changes neither.
Whizi does not publish per model vision support. There is no vision or multimodal capability flag anywhere in the catalogue or the cost index, so Whizi cannot tell you which Llama row reads an image. Image parts are forwarded to whichever model the turn is on and the result is between you and the model.
- Llama is in the catalogue and both priced rows cost 1 credit per message
- Pro is the plan that unlocks the Llama set, and Powerhouse includes it
- Starter includes no Llama model
- A free account can try any text model inside a 7 message lifetime allowance
- A Llama turn runs on the flat 40,000 input and 20,000 output token budget, because both priced rows sit above the 93,000 token threshold
- No Llama row sits on any of the six Auto rungs, so Llama is always a manual pick
- Whizi publishes no per model vision flag, so which Llama row reads an image is not something it can tell you
Frequently asked questions
Which Llama models does Whizi include?
The Pro model set names three Llama identifiers: the two Llama 4 variants plus the Llama 3 workhorse. Two of them are priced in the Whizi Cost Index, both at 1 credit per message, with published context windows of 1M and 131K. The catalogue may carry further Llama identifiers, and anything not named in the Starter or Pro sets requires Powerhouse.
How many credits does a Llama message cost?
One credit, which is the bottom rung of the ladder and the same as one message on the Whizi house model. Starter cannot open a Llama row at all, and a free account can, but only inside its 7 message lifetime allowance, so Pro is the first plan where that 1 credit buys a usable number of Llama turns.
Do I need the top plan to use Llama?
No. Pro is the cheapest plan that opens a Llama row, at $29.99/month. Starter does not reach one, and Powerhouse is needed only for a Llama identifier the Pro set does not name. A free account is not model gated at all, so it can open a Llama row inside its 7 message lifetime allowance.
Can I switch a conversation from Llama to another model?
Yes. Each turn is charged at the rung of the model that answered it, so a thread that moves between families is billed per turn at each rate rather than at the thread level. A model never leaves the catalogue once it has been offered, because a conversation carries the model it was created with, so an old Llama thread still opens later. The mechanics are in switching models mid conversation.
What can I attach to a Llama turn?
Attachments are handled per request rather than per model, and they cost no credits of their own. The caps are 10 MB per file, 4 attachments per prompt and 20 MB in total per prompt. The accepted types are JPEG, PNG, WebP, GIF, PDF, plain text, Markdown, CSV and JSON.
Does web search work with Llama?
Whizi does not publish a per model web search list. Search is a per request toggle with three modes, off, probe and native, rather than a per model capability flag, and the per model matrix was never enumerated, so there is no Llama specific answer to give. The toggle itself is covered in web search.