Compare several models on the same question

Ask VIGPT a question and, by default, Auto reads it and hands it to the best model in your plan. Most of the time that's the whole story — you get a good answer and move on.

But some questions don't have one right answer. The wording of an important email, the framing of a plan, a naming decision, a tricky judgment call — here it can help to see how a few different models handle the same prompt, then pick the one you like best. This is a short guide to when that's worth doing and how to do it.

When comparing is worth it

Reach for a comparison when the answer is a matter of judgment or the stakes are high:

• No single correct answer. Strategy, wording, tone, naming, structure. Different models have different instincts, and seeing two or three gives you options to choose from.
• Quality-sensitive work. A contract clause you'll rely on, a message that has to land right, a plan you're about to act on. A second model is a cheap second opinion.
• The first answer feels off. If a reply is thin or misses the point, asking a different model is often faster than arguing with the first one.
• You want range, not just an answer. Sometimes the point is to see how differently the question can be approached.

When it isn't

For most questions, comparing just spends more of your credit for the same result:

• Straight factual lookups. One correct answer is one correct answer.
• Simple, everyday tasks. A quick rewrite, a short summary, a definition.
• Anything Auto already handles well. Auto biases toward quality, escalates when it's unsure, and takes a second look when a first answer comes back weak. For a lot of work, that's already the comparison you'd have run by hand.

If you're not sure a question needs a comparison, it probably doesn't. Start with Auto and only reach for this when the answer matters enough to be worth the extra minutes.

How to compare, step by step

The idea is simple: keep the question identical and change only the model.

1. Pick a specific model instead of Auto. In the message bar, choose a model rather than leaving it on Auto. Ask your question.
2. Open a separate chat and pin a different model. Paste the exact same prompt. Same words, same attachments — the only thing that changes is which model answers.
3. Repeat for as many models as you want to weigh, one chat each.
4. Read the answers side by side and keep the one that fits.

Keeping the prompt word-for-word identical is the part that matters. If you reword the question between models, you're comparing two questions, not two models.

You can also switch the model inside a single chat and ask again — your history and attached files stay in place. It works, but the second model now sees the first model's answer in front of it, so it's reacting to that answer rather than starting fresh. For a clean, independent comparison, use separate chats.

If your question is about a file, separate chats fit naturally: each conversation keeps one PDF active, so one chat per model means each model reads the same document on its own. If you pin a model that can't read files, VIGPT will prompt you to switch to one that can.

A lighter option: let Auto lean toward quality

If running the comparison by hand is more than you want to do, you can tell Auto what to prefer. Set its preference toward highest quality rather than the fastest answer, and it will reach for stronger models more readily on its own — no second chat required. It won't show you the alternatives side by side, but for many judgment calls it gets you to a strong answer without the manual work.

Keep the cost in mind

Every message uses some of your monthly credit. Simple messages cost little; heavier work on the most capable models costs more. Running the same question past three models costs roughly three times a single answer — worth it for a decision that matters, wasteful for a quick lookup.

A couple of plan notes: the flagship models — the most capable ones, and often the most interesting to compare — are part of the Max plan. Prices and what each plan includes are current at the time of writing; see pricing for live numbers.

In short

Stay on Auto for most things. When the answer is a judgment call or the stakes are high, pin a different model in each of a few separate chats, ask the identical question, and compare. Keep the prompt the same, mind the credit, and let the best answer win.