AI Chat Now Answers With the Best Model

The questions that matter most are rarely the quick ones. Why does this proof step work? Which of these three clauses actually governs the late fee? Turn chapter 7 and my lecture notes into a revision plan that fits around my shifts. Those are the questions where a stronger model is worth the wait, and where a weaker one gives you a confident answer you have to check twice.

Scholarly chat now handles them on stronger models. There are two modes, Auto and Best. Paid plans start on Best, and you can switch to Auto, or to a specific model, whenever you like.

The short version

• Two modes. Auto gives fast answers and is unlimited on paid plans. Best gives the strongest answers, using a premium model.
• Best by default on Premium and Laureate. New chats start on Best. Today Best runs on GPT 6.1 Sol with extra reasoning.
• Pick a premium model yourself. GPT 6.1 Sol, Claude Sonnet 5.5 and Claude Opus 5.5 are in the model menu, alongside the unlimited models.
• A weekly premium usage allowance. Premium models draw from it. The model menu shows how much is left as a percentage, and it resets every Monday at 00:00 UTC.
• Chat never stops. When the week's premium usage runs out, your conversation carries on with Auto until the reset.
• Laureate includes 5× the premium usage of Premium.

What Best is for

Auto is good at quick lookups and follow-ups: define this term, summarize that page, rephrase this paragraph. Best is for the work that takes several steps and a careful read of your material:

• Reasoning through a hard problem in your own source. Paste the exercise, point it at the chapter, and ask for the method, not just the answer. Best works through the steps against the textbook's definitions instead of general ones.
• Comparing documents. Ask which version of a policy, contract or syllabus applies and why. Best reads the relevant parts of each and says where they disagree.
• Planning from real constraints. An exam date, a reading list, the weeks you are travelling: it plans around what you give it rather than producing a generic template.
• Producing finished work. Best does not stop at advice. It writes the document, builds the file, or starts the research and brings back the result.

Answers stay grounded the same way on every model. Chat reads your PDFs, notes, recordings and pages, cites the page or timestamp it used, and says when something is not in your material.

Chat that does the work

Scholarly chat is an agent, not a single reply. For a bigger request it plans, reads your sources, searches the web when it needs to, and can hand longer jobs to a background task, so you can keep talking while it works.

Here is a real example from testing:

Plan a 5-day study schedule for the AWS Solutions Architect Associate exam, then write it into a downloadable calendar file I can import.

On Best, chat replied straight away that it was putting the plan and the calendar together. It started a background task that searched for the current exam domains, built a five-day plan around them, and posted an importable .ics calendar file back into the conversation. A follow-up question about cutting one evening from three hours to two got a specific answer: keep the error review, cut the number of practice questions.

Background tasks run on the same model and reasoning as the conversation that started them. A task you start on Best is done by Best, and it counts toward your premium usage like the rest of the conversation.

How premium usage works

Premium models cost more to run than Auto, and a long, multi-step answer costs more than a short one. A fixed number of messages would treat a one-line definition the same as a two-hour research task. Premium usage is measured by the work each answer actually takes instead:

• A short question uses very little.
• A long thread, a large document or a background task uses more.
• Claude Opus 5.5 uses premium usage fastest, and GPT 6.1 Sol and Claude Sonnet 5.5 use less. The model menu marks each one so you can see the difference before you choose.

You will find the Premium usage meter at the bottom of the model menu. It shows the share left this week and when it resets. When it runs low it turns amber. When it reaches zero, chat answers with Auto until the weekly reset (Monday 00:00 UTC), and the menu tells you so. Nothing you were doing is interrupted.

Auto never draws on premium usage, so you can always switch to Auto for quick questions and save the allowance for the hard ones.

Choosing a model

Open the model menu next to the send button:

1. Auto for fast answers, with unlimited use on paid plans.
2. Best for the strongest answer Scholarly can give. This is where paid chats start.
3. Premium models to pick GPT 6.1 Sol, Claude Sonnet 5.5 or Claude Opus 5.5 by name, for example when you prefer how one of them writes.
4. Unlimited models for the full list of models that never use premium usage.

Plans

• Free: chat on Auto within the free plan's lifetime chat allowance. Best and the premium models are a paid upgrade, and the model menu shows where they unlock.
• Premium ($30/month, or $180/year): chats start on Best, with a weekly premium usage allowance and unlimited Auto.
• Laureate ($99/month): the same, with 5× the premium usage of Premium, for people who work in chat all week.

See pricing for everything each plan includes.

Try it

1. Open chat and attach the document you are working from, or pick it from your library.
2. Ask the question you would usually have to check twice.
3. If the answer should become something, a plan, a calendar, a report or a set of flashcards, ask for that too and let chat build it.

For an overview of what chat can do with your files, see the AI workspace page. If the answer you need is better watched than read, the same source can become an AI video lecture.