The core purpose of TextGPT is to let users use ChatGPT and AI image generation as if they were texting a friend. It does not require installing any app and has no subscription; all interaction is done via text message. According to the official description, users can ask any question and receive fast, accurate answers, generate realistic images from simple text, and send photos for the AI to understand or edit.
In terms of main capabilities, TextGPT runs on OpenAI's GPT-5.6 series models, with training data up to February 2026. Users can choose from three tiers on the account page: Luna is the fastest and consumes the fewest credits, Terra is the default balanced option, and Sol performs deeper reasoning for difficult problems. All three tiers can access the internet through OpenAI's web search, so they can handle both general knowledge and answer events happening in real time. Users can also adjust how long the assistant thinks before replying.
For image generation, users start with the 🖌️ or 🖼️ emoji followed by a description to generate an image, for example "🖌️ surrealist image of a dog park carnival." Images are generated by OpenAI gpt-image-2. Images are billed according to the quality level selected on the account page: draft 1,422 credits, standard 4,242 credits, and highest quality 13,722 credits, and the cost already includes the SMS fee.
Users can also send photos directly. After a photo is sent, TextGPT determines the intent: asking questions about the photo will be answered by text message and billed the same as other questions; requests for modification (such as "make the sky purple" or "remove the background") will return the edited image, with the price determined by the same quality setting.
In terms of usage, users interact with TextGPT via text message. The text messages themselves are billed by segment (160 characters per segment, or 70 characters if emojis are included). Receiving text costs 312 credits, and sending each text segment costs 327 credits, so short answers consume one segment, while long answers may consume multiple segments. The AI part is billed by token, approximately 4 characters or 0.75 words, and the rate depends on the selected tier; the default Luna is about one tenth of Terra. Increasing the thinking time will increase the cost, because thinking time is billed at the writing rate. Web search costs 600 credits each time.
There are two credit packages: $12 gets 240,000 credits, and $24 gets 480,000 credits. At the default tier, $12 is approximately enough for120 questions, depending on answer length and how much thinking is involved. Users can check their balance and change the answer model on the account page.
Use cases include: quickly getting AI answers via text message without installing an app, generating images, editing photos, querying information in real time, and so on. Other features also include using past conversations to improve results, backing up conversations to the cloud, and advanced tuning and configuration.
Regarding data storage, the official description says: if the memory feature is enabled, messages will be stored in a database; if memory is disabled, messages will not be stored. Users can also delete messages at any time.
In terms of limitations, the service relies entirely on the SMS channel, and billing is calculated based on both SMS segments and tokens, which may increase usage costs; image generation and web search incur additional credit consumption. The official materials do not provide information on other platform support, specific available countries or regions, free trials, and so on.
