Models, limits & capabilities
Exactly what powers SproutGPT and what it can and can't do today.
| Everyday chat | DeepSeek V4 Flash (open-weight), fallback gpt-oss-120b |
|---|---|
| Deep Think (Pro) | DeepSeek V3.2 Thinking, for maths, logic and harder problems |
| Serving | Vercel AI Gateway and its hosting providers |
| Context | Last 16 messages of the current chat (up to 8,000 characters each) |
| Answer length | Up to ~2,600 tokens (free) or ~4,000 tokens (Pro) per reply |
| Web search | Yes, live web search with cited sources (5/day free, 200/day Pro), via gpt-oss-120b and Perplexity search |
| Photos | Web app: attach a photo (homework, worksheets, notes), read by Qwen 3.7 Flash, an efficient vision model. iPhone and iPad app: the same, plus the camera (the app is in App Store review) |
| Usage counting | Real tokens processed per exchange (message + context + reply); 1,000 tokens = 1 leaf, max 12,000 tokens charged per exchange |
| Free allowance | About 20 to 40 messages a day (40 leaves); Pro about 400 a day (fair use). Long file or PDF chats use more: often 5 to 12 leaves per message |
| Images | FLUX.2 (Black Forest Labs), with Recraft as a fallback, via Vercel AI Gateway. 3 a day free, 30 with Pro |
| Files & voice | PDF, Word and text files are converted to text for the chat (not stored). Voice uses your device's built-in speech |
| Training on your chats | Never. Conversations aren't kept on our servers, except a chat you choose to share as a link |
Why efficient models?
Most everyday tasks (emails, explanations, plans) don't need the largest frontier model. Efficient open-weight models answer quickly and use less compute per reply. See how much energy AI uses.
FAQ
Which AI model does SproutGPT use?
Everyday chat uses DeepSeek V4 Flash, an efficient open-weight model, with gpt-oss-120b as a fallback. Pro Deep Think uses DeepSeek V3.2 Thinking. Models are served through Vercel AI Gateway.
Can SproutGPT browse the web or see images?
Yes to both. Questions about anything current (news, prices, weather, scores) trigger a live web search through Perplexity's search API, and answers show numbered source cards. Free accounts get 5 searches a day, Pro up to 200. It can also read a photo you attach (via Qwen 3.7 Flash, an efficient vision model).
How much of a conversation does SproutGPT remember?
It reads the most recent 16 messages of the current chat (each up to 8,000 characters). Older messages in very long chats are not sent to the model.
Start growing your first tree 🌱
Free every day, no sign-up. Your first answer is a few seconds away.