Which AI Model Should You Pick in CoPilot Interview? A Round-by-Round Guide

Pick Mercury 2.5 for behavioral rounds and quick conversational answers, GPT-6 Luna for coding and anything that is on screen rather than spoken, and Claude Haiku 5.5 or Gemini 3 Flash for system design and long technical answers. That is the short version of which AI model is best for an interview assistant. Which ones you can pick depends on your plan: Mercury 2.5 by Inception is the default on Free, GPT-6 Luna comes with Pro and also reads screenshots, and Pro+ adds Claude Haiku 5.5 and Gemini 3 Flash. On Standard and above, Dual AI Mode shows two models side by side so you do not have to choose.

The model map by plan

Prices are as shown on our pricing page on October 25, 2026. Every plan uses the same live loop: captions of the question, then a structured answer in about 2 seconds, with shorter, longer and redo controls.

Behavioral and conversational rounds: Mercury 2.5

Behavioral questions reward speed and structure more than depth. You need a situation, an action and a result in the time it takes the interviewer to finish asking. Mercury 2.5 is fast and produces that structure well, which is why it is the default and why the free plan is enough for most behavioral practice.

Use the shorter control if the answer runs long; in a live call, a 40-second answer you can say beats a complete one you cannot.

Coding and on-screen problems: GPT-6 Luna

Coding rounds are different because the question is often not spoken. The interviewer pastes a problem into a shared editor or shows a SQL table. GPT-6 Luna, on Pro, reads screenshots: press Ctrl+Shift+C to turn Coding mode on and Ctrl+Shift+S to read the problem from the screen. Coding mode returns code with Big-O and edge cases; Ctrl+Shift+E answers the next question in words while Coding stays on, for the "walk me through it" that follows.

Our comparison of coding versus conceptual questions explains why the two need different handling.

System design, case and long technical answers: Claude Haiku 5.5 and Gemini 3 Flash

System design and case questions are multi-part: clarify, propose, trade off, estimate. We point people to the two Pro+ models for these because they hold a longer structure across follow-ups. The honest caveat: a longer answer is not a better one in a live call. Read the first two lines, speak, and come back for the trade-offs when the interviewer asks.

For a case interview, the same applies with numbers: use the model's framework, and do the arithmetic yourself, out loud.

Use Dual AI Mode instead of guessing

On Standard and above, Dual AI Mode shows answers from two models side by side. In practice this is how most people settle the question: run two for a week of mock interviews and notice which one you actually read. Our page on assistants with two AI models covers the pairings that work, and which models power AI interview assistants compares what other tools run.

How to test models before the interview

1. Start an AI mock interview with the real job description and your resume.
2. Pick three questions from the round you are worried about and answer them with Mercury 2.5 selected.
3. Repeat the same three with the other model you have access to, or run both in Dual AI Mode.
4. Keep the model whose answer you could say out loud without editing.

Do this a day before, not an hour before, and follow your employer's rules on whether live help is allowed at all.

Where CoPilot Interview fits

CoPilot Interview is a desktop AI interview assistant that lets you choose the model rather than locking you to one. Start on the free plan with Mercury 2.5 and the mock interview; move up only if a coding or design round needs it. Disclosure: CoPilot Interview is our product, so weigh what we say about it accordingly.

Frequently asked questions

Which AI model is best for an interview assistant?

It depends on the round. Mercury 2.5 is fast and structured, which suits behavioral questions. GPT-6 Luna reads screenshots and returns code with Big-O and edge cases, which suits coding rounds. Claude Haiku 5.5 and Gemini 3 Flash hold long, multi-part structures for system design. On Standard and above, Dual AI Mode shows two side by side.

What model does the free plan of CoPilot Interview use?

Mercury 2.5 by Inception is the default on the Free plan, with 15 minutes of live help every day in sessions of up to 15 minutes, and AI mock interviews included. It suits behavioral and conversational questions. GPT-6 Luna requires Pro; Claude Haiku 5.5 and Gemini 3 Flash require Pro+.

Which model reads screenshots of a coding problem?

GPT-6 Luna, available on the Pro plan. Press Ctrl+Shift+C to turn Coding mode on and Ctrl+Shift+S to read the problem from the screen. Coding mode returns code with Big-O and edge cases, and Ctrl+Shift+E answers the next question in words while Coding mode stays on.

What is Dual AI Mode?

Dual AI Mode shows answers from two AI models side by side for the same question. It is included on Standard, Pro and Pro+. It is the practical way to compare models during mock interviews and settle on the one whose answers you would say out loud without editing.

Do I need Pro+ for system design interviews?

No. Any model will produce a system design framework, and the free Mercury 2.5 is enough to practice the structure. Pro+ adds Claude Haiku 5.5 and Gemini 3 Flash, which we point people to for long, multi-part answers. Test with a mock interview before paying for a round you may not need.