The short answer
A model is unavailable in Whizi for one of three reasons: your plan does not include it, the model id did not resolve to a usable model in the catalogue, or the provider behind that model failed on that request. The error text tells you which of the three you hit.
| Cause | What you see | What fixes it |
|---|---|---|
| Plan gate | A 403 or a 402 | Upgrade, or pick a model your tier includes |
| Model id | A 400 saying the model is not available or not supported | Pick a model from the picker rather than reusing an old id |
| Provider failure | A 429 or a 502 mentioning the model provider | Retry, or send the same message on another model |
Every backend error arrives as JSON in the shape {"error": {"code": "...", "message": "..."}}, and the website shows the backend message verbatim, so the string on your screen is the string in the table below.
Plan gate errors, word for word
These mean the account is real and the model is real, but the tier does not reach it. Access is resolved on the server by model identifier, so this is an entitlement rather than an interface bug.
| Message | Status and code | Trigger |
|---|---|---|
Upgrade your plan to use this model. | 403 tier_upgrade_required | A paid account picked a text model its tier does not include |
This model is only available on the Powerhouse plan. | 403 powerhouse_required | A creative model on the Powerhouse-only list was requested from a lower tier |
Choose a subscription to use Whizi. | 402 subscription_required | The route requires a paid plan and the account is not subscribed |
Your free messages are used up. Start a subscription to keep chatting. | 402 free_limit_reached | A free account is already at its lifetime message cap |
The free tier lifetime allowance is 7 messages, counted for the life of the account with no reset path. Once it is spent, every send returns the 402 in the last row above.
Which plan opens which models
Tier rank runs free 0, Starter 1, Pro 2, Powerhouse 3, and a higher tier always satisfies a lower requirement, so each tier includes everything below it.
| Plan | What it opens |
|---|---|
| Free | A lifetime allowance of 7 messages with no reset. After that, Your free messages are used up. Start a subscription to keep chatting. |
| Starter, $15.99/month | Exactly four picker entries: Auto, Whizi AI, one OpenAI GPT model, and one fast Google Gemini model |
| Pro, $29.99/month | Starter's four plus 37 further model identifiers, so 41 listed identifiers |
| Powerhouse, $49.99/month | Everything else in the 280+ catalogue |
The house model Whizi AI sits at Starter tier, so the default assistant is never paywalled above Starter. The tier fallback is Powerhouse rather than Pro: any identifier the catalogue gains lands on the top tier until it is deliberately promoted into the Pro list. A model that appeared this week and sits on Powerhouse is usually not a mistake.
Pro is a curated set rather than everything cheap: one or two current flagships per family, each family's fast tier, and the high-volume workhorses. Claude Sonnet is the Pro ceiling and the Claude Opus class is Powerhouse. Agent personas are a Pro feature gated by identifier ahead of the Powerhouse fallback, so they do not inherit the top-tier default.
The reliable way to avoid this class of error entirely is Auto. Auto is available on every paid plan, it is a router rather than a model, and it only ever picks from models the subscriber can already reach, never above the tier.
The model is missing from the picker
Check which section you are looking at before assuming it is gone. The picker has six: Recommended, Agents, Image Generation, Video Generation, Audio Generation, and All models. Recommended is deliberately constrained to models reachable at Pro or below, so the top of the picker is never Powerhouse-locked. A model that needs Powerhouse will not be in Recommended.
If you see image models where you expected text models, the composer is in creative mode, which switches the picker to image models and swaps the placeholder to an image prompt. Media generators are not gated per model at all: they skip the per-model gate and are tiered by picker section instead.
A model that was there before still answers. A model never leaves the catalogue once it has been offered, because a conversation carries the model it was created with. The identifier keeps resolving even when the model behind it has been replaced: a chat pinned to a retired model is silently answered by its replacement rather than refused, so an old thread still opens. The picker contents are served by the Whizi backend at GET /api/models, and the lists hardcoded into the web and mobile clients are offline fallbacks only, so what you see is the live catalogue rather than a list baked into your browser.
Model id errors, word for word
These fire before any provider is called. They mean the id in the request did not resolve to a model that can do the thing being asked.
| Message | Status and code | Trigger |
|---|---|---|
A model is required. | 400 invalid_model | The model field is missing or blank |
That model is not available for text chat. | 400 unsupported_model | The requested model id does not resolve to a text model in the catalog |
A title and supported text model are required. | 400 invalid_conversation | A conversation was created with a missing title, or a model that is neither a text nor a creative model |
Choose Whizi Image, Nano Banana, FLUX, or Stable Diffusion. | 400 invalid_image_model | The requested model is not an image generation model |
Choose Whizi Music AI to generate a song. | 400 invalid_music_model | A music job named something other than the music model |
Choose Whizi Video AI to generate a video. | 400 invalid_video_model | A video job named something other than the video model |
That model is not available for text chat. is the one people hit most, and the usual cause is asking a generation model for prose. The Image Generation section is Whizi Image, Nano Banana, Flux and Stable Diffusion, Video Generation is one row, and Audio Generation is one row. None of them answer text turns. Pick a text model from Recommended or All models and resend.
Provider failures, where retrying is the fix
These are not your account and not your plan. The model exists, your tier reaches it, and the upstream provider failed on that request.
| Message | Status and code | Trigger |
|---|---|---|
The model is temporarily rate-limited. Please retry shortly. | 429 provider_rate_limited | The upstream provider answered the connect with a 429 |
The model provider rejected the request. | 502 provider_error | The provider's error body could not be parsed, or parsed to an empty message |
The model provider returned an empty stream. | 502 | The upstream response was OK but carried no body |
The model provider returned an error. | In-stream error event | The provider sent an error frame mid-stream with no message of its own |
The model stream was interrupted. | In-stream error event | Any other mid-stream failure |
Error generating response. Please try again. | 502 provider_error | An image, video, music or speech generation failed for a reason other than a provider 429 |
No setting on your side clears any of these, and Whizi publishes no retry window. Resend in a moment, or send the same message on a different model in the same conversation. Switching model does not lose the thread.
Two more strings belong to the same family. Failed to send message. is the website's fallback toast when a send fails and the backend supplied no message at all, which usually means a dropped connection rather than a model problem. Sorry, there was an error generating the response. is written into the assistant bubble when a stream ends with no final message and no accumulated content. In the side-by-side compare view, a failed side reads The model did not respond. Please try again.
Errors that look like model errors but are not
Three strings that name no model get read as the model being broken: Your monthly message limit has been reached. and Your weekly message limit has been reached. are HTTP 429 message_limit_reached, Too many requests. Please wait and try again. and Hourly message limit reached. Please try again later. are HTTP 429 rate_limit_reached, and Generation already in progress. Please stop it or wait for it to finish. is HTTP 409 generation_in_progress. None of them is a model gate, and switching model fixes none of them.
The allowance side is covered in out of credits, and the rate limit and concurrency side in a message will not send.
- A 402 or 403 means the plan, a 400 means the model id, a 502 means the provider, and a 429 means either the provider or your own allowance depending on the code
- The free lifetime allowance is 7 messages, and it never resets
- Starter opens exactly four picker entries, Pro opens 41 identifiers
- Anything new in the catalogue defaults to Powerhouse until it is promoted
- Auto never picks a model above your tier, so it never triggers a plan gate
- Recommended is limited to models reachable at Pro or below
- A model never leaves the catalogue once it has been offered
- For provider failures the only fixes are retry or switch models
Frequently asked questions
Why is a model greyed out in Whizi?
Because your tier does not reach it. Sending anyway returns Upgrade your plan to use this model. as HTTP 403 tier_upgrade_required, or This model is only available on the Powerhouse plan. as HTTP 403 powerhouse_required for a creative model on the Powerhouse-only list. Upgrade, or pick a model your tier includes.
What does "That model is not available for text chat." mean?
Almost always one of two things you did without noticing. Creative mode was left on, which points the picker at image models, so the row you picked was an image row. Or the chat was started on a row from Image Generation, Video Generation or Audio Generation and later used for prose. Turn creative mode off, choose a text model, and resend.
How long do I wait after "The model is temporarily rate-limited. Please retry shortly."?
Whizi does not publish a wait time, so any number would be a guess. That string is HTTP 429 provider_rate_limited and it means the upstream provider refused the connect, not that your account did anything wrong. Resend in a moment, or send the same message on a different model in the same conversation, which keeps the thread and its context intact.
A model disappeared from my model picker. Was it removed?
Almost certainly not. Start with which of the six sections you are in, because the picker splits into Recommended, Agents, Image Generation, Video Generation, Audio Generation, and All models, and Recommended is limited to models reachable at Pro or below. If the rows in front of you are image models and the composer placeholder is asking for an image prompt, creative mode is on and the picker has switched with it. Turn it off, then open All models.
How many messages do I get before Whizi asks me to subscribe?
Seven. The free lifetime allowance is 7 messages, counted for the life of the account with no reset path, so it is not a monthly free tier. After that, sends return Your free messages are used up. Start a subscription to keep chatting. as HTTP 402 free_limit_reached, and no amount of waiting clears it. Starter at $15.99/month opens four picker entries, including Auto and the house model Whizi AI.
A model I could use last month is locked now. What changed?
Model access is resolved per request by model identifier, through the same lookup every time: the Starter set, then the agent set, then the Pro set, and otherwise Powerhouse. So the answer can move for two reasons. Your tier is one. The other is the catalogue: any identifier it gains lands on Powerhouse until it is deliberately promoted into the Pro list, so a row that looks familiar but is new to the catalogue sits on the top tier by default. See switching models mid-conversation.