Guide · Multi Chat

See more sides of the answer.

Multi Chat lets you ask several AI models the same question. Compare their responses, examine the differences, and decide what deserves a follow-up.

In this walkthrough, we plan a family paper-bridge challenge. You can use the same steps with a question of your own.

Shows the web app · Last reviewed 1 October 2026

What you can do

Put one question to several AI models and read their answers side by side.
Ask the models to critique one another, see where they agree, and bring their answers together in one.
Keep going with every model or just one, or carry chosen results into a focused discussion.

Before you begin

Sign in to an adult account and choose Multi Chat in the switcher at the top of any chat.
The lineups available to you depend on your plan. The free trial runs Basic. Mega (OpenAI · Anthropic · Google), Medium and Wildcard (Grok · DeepSeek · Qwen) are part of Plus, and show “Part of Plus” on a trial account.

1Choose a lineup and ask your question#

Choose a lineup, then type your question. A preset gives you a starting selection of models; choose Custom if you want to pick them yourself. For our example:

We have two hours for a family paper-bridge challenge using paper, tape, and coins. Suggest three approaches, explain the trade-offs, and give us a simple plan.

Check the estimated cost of the round before sending. Once there is enough recent data, the typical time a round takes (for example, “~6s”) appears next to it. The final cost can differ from the estimate.

Synthesize is switched on by default, so a combined answer arrives automatically after the models finish. You can switch it off before sending.

Saved memories can come with your question: pick them with Memories beside the message box, and each model in the round gets the same ones.

Multi Chat before the question is sent. The Lineup row offers Mega, Medium, Basic, Wildcard and Custom with Mega selected, a per-round estimate and a typical round time beside it, and the models the lineup includes; the question is typed in the box beside the Run button, and a line under the box says Auto Synthesize is on and what it adds per round. — Choose a lineup and check the estimate before sending your question.

2Compare the answers#

After you send, each selected model answers the question. On a larger screen the responses appear in columns; on a phone, you move between tabs.

Look for a useful difference. Did one answer notice a practical constraint? Did another suggest a simpler starting point? You do not need to choose a winner to find something worth exploring.

A finished Mega round on a computer: the question at the top, answers from OpenAI GPT-6 Astra, Anthropic Claude Opus 5 and Google Gemini 3.1 Pro Preview in side-by-side columns with the round total, and buttons below for Reply to all, Re-run with other models, Cross-review, Bearings, Synthesize, Discuss these results, Save as PDF and Share round. — Compare the different approaches to the same question.

A follow-up round on a phone: the earlier round folded as Previous round, the new question, a tab for each model with OpenAI GPT-6 Astra selected and its answer below, and a row of round actions above the message box. — On a phone, each model's answer has its own tab.

3Ask the models to review one another#

Choose Cross-review to ask each model to critique the other models’ answers. A model never reviews its own. The critiques arrive as a new round, and the original responses stay where they were, so you can read both.

A critique can help you spot an omission or question an assumption. It is still an AI response, so assess the reasoning yourself.

The Cross-review tab of a Mega round: a Cross-review round with three columns, one critique each from OpenAI GPT-6 Astra, Anthropic Claude Opus 5 and Google Gemini 3.1 Pro Preview, each marked complete, saying what worked in the other models' answers and where the reviewer disagrees. — Ask the models to examine one another's answers.

4Find where the answers agree#

Choose Bearings for an agreement map of the finished round. For each claim, it shows which models agreed, which only partly agreed, and which disagreed or stayed silent, with quotes from their answers.

Bearings describes the answers already given; it does not ask the models again. Cross-review asks them for new critiques. Neither is independent fact-checking, and models can share the same mistakes.

Bearings for the round, below the Compare, Bearings, Synthesis and Cross-review tabs: a short summary, then each claim with a chip per model showing whether that model agrees, partly agrees or stayed silent, and the ideas the models raised. — See where the answers agree and where they differ.

5Bring the perspectives together#

Because Synthesize is on by default, the round’s combined answer appears once every model has finished. It stays in the conversation, so you can come back to it. Read it with the original responses in mind, especially where they disagreed.

It can be a useful place to find a starting plan for the paper-bridge challenge. It does not mean every difference has been resolved.

The Synthesis tab selected beside the Compare tab, showing one combined answer from OpenAI with sections for key points of agreement, meaningful disagreements and unique insights, each naming OpenAI, Anthropic or Google. — Bring the perspectives together in one answer.

6Continue the question#

Use the main follow-up box at the bottom to ask every model another question. Each model continues its own thread. For example:

Which approach uses the least tape, and what should we test first?

Use an answer’s own reply control when you want to continue with just that model. Earlier rounds fold up above the new one; open any of them to read it again.

A follow-up round in the same conversation: the first round folded at the top as Previous round with an Expand button, and the new question with fresh answers from the same three models below it. — Continue the question with the models you chose.

7Open a focused discussion#

Choose 💬 Discuss these results…, tick the answers and results you want to talk about, and start the discussion. It opens as a new, linked chat, so a particular idea gets its own conversation while the original round stays as it was.

The picker shows roughly how much of the chosen model’s context your selection uses, so nothing is cut off without you knowing.

The Debrief picker over the conversation: a Find in results box, a checklist for each round listing every model's answer, Synthesis and Bearings, with the first round's Anthropic Claude Opus 5 answer and Bearings and the second round's Synthesis ticked, then a model menu, a size estimate, a first question and a Start button. — Choose which results to carry into a focused discussion.

The new linked Single Chat. A label at the top reads Discussing results from, followed by the Multi Chat's title, and the first message, formatted as headings and lists, carries the chosen rounds and answers, starting with the question and Anthropic Claude Opus 5's answer. — Discuss selected results in a linked conversation.

Choices and boundaries

Cost

Every model in a round is a separate answer, so a round costs roughly the sum of its models. The estimate is shown before you send and the actual cost once the round lands. Cross-review asks every model again, so it adds about another round; its estimate appears beside the button. Pressing Synthesize by hand runs, and charges for, another synthesis.

Carrying results forward

The Include in next question boxes add the latest synthesis or Bearings to your next question as context, so the models see what was concluded without running the step again.

Slow or failing models

If a model is slow, you can skip it and the round finishes with the answers that arrived. If a model fails, its column says so; the others still answer.

Sharing a round

Share round creates a public, read-only page for one round. It needs at least two completed answers and carries the models’ answers, never the Cross-review critiques or Bearings. Anyone with the link can read it until you stop sharing, and a copy someone already saved can’t be taken back. Encrypted chats can’t be shared.

Children

A child’s Multi Chat uses their own Basic and Medium lineups, built from models cleared for children; there is no Mega-class lineup for children. Children can’t share rounds, and Discuss these results is for adults only. If a parent holds each of a child’s messages for approval, Multi Chat isn’t available to that child; they use Single Chat instead. The Safety filter applies in Multi Chat as it does everywhere else.

Try it

Bring a question with more than one reasonable approach. Compare the answers, inspect one difference, and ask one follow-up.

Open Multi Chat

Next: Memory explains how True North keeps useful context and adds it to your questions, in Multi Chat as in Single Chat, and Sharing & Gems keeps a passage from any answer, critique or synthesis in your Gold Chest.

Exit supervised mode