The first minutes with a new tool decide whether you come back. This update is about those minutes: where a new account lands, what happens when your first choice is not on your plan, and how ToRun keeps up when you ask for more than one thing at once.
Start in the chat
When you create an account on the web, you now land in the chat instead of a welcome wizard. Suggestions above the message box, such as writing an email or planning your week, fill in the box so you can change them first; nothing is sent until you send it. The welcome steps, tips and the in-app assistant's pop-up wait until you have your first answer, and then ToRun offers a short tour.
Try it on your phone, without an account
The ToRun app lets you try ToRun without an account once your phone passes a quick device check. When that check could not be completed, for example because it got no answer, you used to see only examples. Now you can send a few chat messages instead. Voice, songs and cartoons still need a phone that passes the check, or an account.
An answer, whichever model you pick
A new account often starts by picking a model or Deep Research that its plan does not cover yet. That reply used to end empty, with a button to answer with Auto. Now Auto answers on the same reply, which opens with one line saying what your plan did not cover and which model answered, and you are billed only for the model that answered. In the ToRun app, a model your plan does not cover says so on its row in the model list.
Your language, your chat
ToRun answers, and names your chats, in the language you write in, even when the app is set to another language. When you ask for several pictures in one message, they are made at the same time, and each appears as soon as it is ready. Voice mode knows the chat you came from, the pictures in it and your last messages. In the ToRun app, sharing photos to ToRun offers to add them to the chat you just had open, and the full-screen image viewer can save a picture or share it.
Long pages are not lost
When the AI writes a long page into a Canvas and its answer is cut off partway, what was already written is now saved as your Canvas, and the AI goes on to finish it. Every reply has a length limit sized for the job, so a single answer can no longer run on for many minutes and cost far more than you expected.
Also in this update
- When you choose Auto, ToRun now picks only from models we have measured for that kind of work. A newly released model or a new provider is not used by Auto until we have measured it; you can still pick it yourself.
- The venue assistant recognizes the language of more questions, such as a short English question on a page in another language or a question in Kazakh, and answers in that language, even after a long conversation in another language. When a question is written in another language than the page, the directions in the answer now come in the question's language too, for the languages venue pages are offered in.
- When the venue assistant's answer names a stand, for example at a fair, the page can now show that stand on the venue's plan, even when the question did not name a place. Asking to see it on the map or the plan shows the place the conversation was last about. When no place has come up yet, the assistant asks which one is meant, and for a venue without a plan on its page it says so.
- Before the venue assistant answers, the page runs a short security check that tells people apart from automated programs. In some browsers that check never appeared or never finished, and the visitor was told they could not be verified, without having done anything wrong. Now the visitor can still ask their question in a short conversation, and the page says so once. No IP address or other personal data is kept for this.