Choosing an AI Model

Scholarly lets you choose which AI model powers your conversations, slide decks, and video lectures. Different models have different strengths -- swapping between them is a good habit when one output does not quite hit the mark.

Where You Can Pick a Model

You can pick an AI model in several places:

• Home chat box -- The model picker now sits right next to the Home chat input, so you can swap models before sending your first message — no need to open a chat first.
• AI Chat -- Open the model menu above the chat input to switch models for a new or existing conversation.
• AI Slides -- Pick a model in the slide creator before you click Generate.
• AI Video Lectures -- Pick a model in the video creator before you click Generate.
• Flashcards (paid plans) -- Pick which model writes the deck in the customize step of the flashcard creator. Free users generate with DeepSeek V4 Flash; paid plans default to Gemini 3 Flash, with GPT 5.4 Mini and Mistral Small 4 as alternates.
• Worksheets, Study Guides, Mind Maps, Infographics (paid plans) -- Pick a model in each create modal before generating. Free users see the option with a lock badge.

Each provider's logo (OpenAI, Google, xAI, Mistral, Anthropic, DeepSeek, Moonshot AI) shows next to its models in the picker so you can tell families apart at a glance. Your last selection is remembered per feature -- changing the model for video lectures doesn't change your chat default, and vice versa.

Recommended Models and Other Models

Every model picker — in chat and in every AI create tool — is organized into two groups. A short Recommended Models list sits at the top with the best picks for that feature, and everything else lives in an Other Models submenu one tap away. Each model has a short, plain-English description underneath ("fast and lightweight", "clear, well-paced lectures") so you can tell at a glance what each one is good for without having to memorize names.

If you're on a paid plan, you automatically get the recommended model for whatever you're creating — no extra setup needed. If you'd like a different style, open Other Models and pick one.

Browsing the Picker on Free

Free users can open the model picker and browse every available model, with the same Recommended badge and descriptions paid users see. You can read about each model before deciding whether to upgrade. Actually choosing a non-default model stays part of the paid plan, but nothing in the picker is hidden.

Available Models

The exact list updates as new models are released. Common options include:

• GPT 5.4 -- The recommended premium OpenAI model. Good at following instructions and producing well-structured writing. Best for summaries, study guides, and structured explanations.
• GPT 5.4 Mini -- Faster and lighter. Great for quick slide decks or short video lectures when you care more about speed than deep reasoning.
• Gemini 3.1 Pro Preview -- Google's newest premium reasoning model for chat. A strong pick for careful explanations, long-context source work, and problems where you want thinking built in.
• Gemini 3.5 Flash -- Google's premium reasoning model. A strong pick when you want careful explanations, long-form analysis, or source-heavy work.
• Gemini 3 Flash -- Fast, capable, and efficient. The recommended pick — and the paid-plan default — for AI Slides, AI Video Lectures, infographics, and flashcards.
• DeepSeek V4 Flash -- A fast, lightweight text model and the free-plan default across generation surfaces (slides, video lectures, infographics, flashcards, research). On a paid plan it's a quick alternative in Other Models when you want speed without giving up quality.
• Kimi K2.7 Code -- Moonshot AI's vision-capable model, available in the AI Slides picker. A good alternate when you want a different take on a deck's layout and structure.
• Grok 4.3 -- Creative, conversational, and confident. Good for brainstorming, essay outlines, and casual explanations. In AI Chat it also offers an Instant mode for fast replies when you do not need deep reasoning -- turn the Thinking toggle off to use it.
• Mistral Small 4 -- A fast European model from Mistral AI with a large context window and optional thinking mode. Good for everyday chat and flashcard generation when you want a different style.
• Mistral Medium 3.5 -- Mistral's stronger reasoning model for paid chat. Use it for complex explanations, multi-step work, and source-heavy questions.
• Mistral Large 3 -- Mistral's larger premium chat model. A good alternate when you want a careful Mistral answer without turning on thinking.
• Claude Sonnet 5 -- Anthropic's newest Sonnet model. Strongest at long-document analysis, careful reasoning, visual source understanding, and nuanced explanations. Pick it when you want the model to really think through a problem.
• Claude Haiku 4.5 -- Anthropic's fast everyday model. Lighter and quicker than Sonnet, still strong at writing and reasoning. Good default when you want Claude's style without waiting.

Locked Models on the Free Plan

Some models require a paid plan. When premium choices are shown on the Free plan, they use a small lock badge so you know what an upgrade unlocks. Selecting a locked model prompts you to upgrade rather than switching to it.

Free models stay fully available without a lock badge. To unlock the rest, upgrade to Ultimate — see Plans and Limits.

You can still browse every model — including locked ones — to compare. The picker doesn't hide anything from you.

How to Pick

A few simple heuristics:

• Need a quick answer, outline, or short video lecture? Start with DeepSeek V4 Flash, Gemini 3 Flash, or GPT 5.4 Mini.
• Need deep reasoning or step-by-step math? Try Gemini 3.1 Pro Preview, Gemini 3.5 Flash, or a reasoning-enabled GPT model.
• Want creative phrasing or brainstorming? Try Grok 4.3 or Mistral Small 4.
• Reading a long PDF or working through a careful explanation? Try Claude Sonnet 5 -- it handles long documents and nuanced reasoning especially well.
• Want a European model family? Try Mistral Medium 3.5 for reasoning or Mistral Large 3 for a premium non-thinking pass.
• Want Claude's writing style without the wait? Try Claude Haiku 4.5 -- fast, lightweight, still thoughtful.
• Not sure? Start with GPT 5.4 Mini for speed, or GPT 5.4 on a paid plan when you want the strongest OpenAI reasoning.

If the first result does not feel right, regenerate with a different model. Slide decks and video lectures are quick to re-run, and each model gives a noticeably different style.

Thinking Toggle (Chat Only)

In AI Chat, you can enable a Thinking toggle in the model menu. When turned on, the model reasons through the problem before replying -- which produces better answers for complex questions, math, and multi-step problems.

• Off -- Fastest responses, no visible reasoning step.
• On -- The model thinks before answering. You see a live progress summary while it reasons, which collapses into a "Thought for X seconds" block when done.

Not every model supports thinking. Some models always reason and do not show the toggle because thinking is built in. Gemini 3.1 Pro Preview always reasons in chat. Both Claude models support the Thinking toggle -- it is especially useful on Claude Sonnet 5 for hard, multi-step problems. Grok 4.3 supports it too: leave Thinking on for careful answers, or turn it off for Instant replies when speed matters more than depth. Mistral Small 4 and Mistral Medium 3.5 also support thinking mode; Mistral Large 3 appears as an instant-response premium option.

Model Data Handling and Privacy

Scholarly never sells your content and never uses your prompts or sources to train AI models. Your material is used only to generate the result you asked for.

The models are run through their AI providers (OpenAI, Google, Anthropic, xAI, Mistral, DeepSeek, Moonshot AI). A few of the most advanced, top-tier models carry extra provider requirements: the provider may hold the request for a short period (typically up to 30 days) for safety and abuse monitoring before deleting it, and may occasionally decline a request that trips its safety systems. The standard models don't carry these requirements — so if you'd prefer to avoid any provider-side retention, pick one of them.

For the full details on how Scholarly handles your data, see the Privacy Policy.

Tips

• Experiment! The fastest way to find your favorite is to generate the same slide deck or short lecture with two different models and compare.
• Some chat models require a paid plan — those show a lock badge in the picker on the Free plan. Model choices inside AI Slides and AI Video Lectures follow those features' normal plan limits.
• For creative or exploratory work, a different model can unlock a completely different angle on the same source material.