Which AI model to choose in VocaLingo Chat
Auto picks a model for each message. You can also choose one of 12: GPT 6, Claude, Gemini, DeepSeek, Grok, MiniMax. All are open to everyone and accept files.
Updated November 2, 20264 min read
If you are not sure which model to take, leave Auto. Pick a model yourself when you want a specific one: the strongest for a hard task, or the cheapest for simple questions.
How to switch the model
Open the Chat tab.
Tap the model name under the VocaLingo AI title at the top of the screen. Another way: ⋮ → Choose model.
In the Chat settings sheet, stay on the Text tab and tap a model. The Images tab chooses the model that draws pictures.
The next message goes to the new model. You can switch in the middle of a chat: the conversation stays, and each answer shows which model wrote it.
Models in the picker
The picker marks each model with a level (Fast, Universal or Maximum) and a price mark from $ to $$$. The descriptions below are the ones the app shows.
| Model | Level | Price | Good for |
|---|---|---|---|
Auto | chosen per message | depends on the model | When you don't know what to pick |
GPT 6 Astra | Maximum | $$$ | Hardest work: coding, research and long tasks |
GPT 6 Sol | Universal | $$ | Most everyday tasks: writing, ideas, analysis |
GPT 6 Luna | Fast | $ | Fast answers to simple questions |
Claude Fable 5.1 | Maximum | $$$ | The strongest Claude: complex projects and code |
Claude Opus 5.5 | Maximum | $$$ | Almost Fable quality, for less |
Claude Sonnet 5.5 | Universal | $$ | Balance for code, writing and agents |
Gemini 3.1 Pro | Universal | $$ | Long documents, images and big chats |
Gemini 3.8 Flash | Universal | $ | Fast and cheap, quality close to the flagships |
DeepSeek V4.1 Pro | Universal | $ | Harder code and reasoning. It is DeepSeek V4.1 Flash with thinking turned on |
DeepSeek V4.1 Flash | Universal | $ | Fast and cheap: code, drafts, images |
Grok 4.7 | Universal | $$ | Up-to-date knowledge and code |
MiniMax M3 | Fast | $ | Fast and affordable, understands images |
Every model is open to every user, Premium is not required. Every model accepts files. If a model is temporarily turned off, it shows the Unavailable badge in the picker: choose another one or Auto.
How Auto chooses
Auto looks at each message and sends it to a suitable model. Under the answer you see which one it was, for example “Auto → Gemini 3.8 Flash”.
Ordinary questions go to Gemini 3.8 Flash.
Complex tasks go to Claude Sonnet 5.5.
CSV and Excel files, audio and code go to GPT 6 Luna.
Very long chats go to MiniMax M3 or Grok 4.7.
Which models search the web themselves
GPT 6, Claude, Gemini and Grok have their own built-in web search. DeepSeek and MiniMax search through VocaLingo's search tool. For you the result is the same: the AI decides when to search, and each search adds a small charge. More in web search and deep research.
How much a message costs on each model
The price of an answer depends on how much text the model reads (your message, the whole chat, attached files) and how much it writes. Models count text in their own units, also called tokens; do not confuse them with VocaLingo tokens on your balance. The table shows how many VocaLingo tokens a million of these units costs.
| Model | VocaLingo tokens per 1M units read | per 1M units written |
|---|---|---|
GPT 6 Astra, Claude Fable 5.1 | 1200 | 6000 |
Claude Opus 5.5 | 480 | 2400 |
GPT 6 Sol, Claude Sonnet 5.5 | 240 | 1200 |
Gemini 3.1 Pro | 240 | 1440 |
Grok 4.7 | 240 | 720 |
Gemini 3.8 Flash | 90 | 450 |
DeepSeek V4.1 Pro and Flash, MiniMax M3 | 36 | 144 |
GPT 6 Luna | 12 | 60 |
The price for Gemini 3.8 Flash is valid until December 31, 2026. On GPT 6 and Grok very long chats cost more per message. The longer the chat, the more each new answer costs, because the model rereads the whole conversation. A new chat for a new topic saves tokens.
Frequently asked questions
Which model is best for translating text?
For everyday messages Auto or Gemini 3.8 Flash is enough. For a long or difficult document, such as a contract, take GPT 6 Astra, Claude Fable 5.1 or Claude Opus 5.5.
An old chat shows a model that is not in the picker. Is that a bug?
No. Each answer keeps the name of the model that wrote it. When a model is replaced by a newer one, old answers still show the old name, and new messages go to the models in the current list.
Does voice mode use the model I chose?
No. Voice mode always works on its own voice model, whatever is selected in the picker. See voice mode.
Contact support
Describe the problem and we will reply by email. Add the file name or job ID if you have one.