Choosing an AI Model

Creation tools start on Auto unless you have saved another choice. In AI Chat, paid members with premium usage remaining start on Best; Free starts on Auto. You can keep those defaults or open the model picker for more control.

What a model costs

The model you choose decides how much a creation weighs against your weekly AI Creation Credit allowance. Every row in the picker carries its badge, so you always see the cost before you choose:

Cost	Models
1 credit	Auto (Free), Auto (Premium), and every named model without a badge, including GPT 6 Luna, Claude Haiku 5.5, Gemini 3.8 Flash, Grok 4.7, Gemini 3.1 Pro Preview, Mistral, GLM, DeepSeek, and Muse Spark 1.3
2× credits	Kimi K3, Claude Sonnet 5.5, Claude Opus 5.5, and Auto (Best)
3× credits	GPT 6.1 Sol, GPT 6 Astra
4× credits	Gemini 4 Argon, Muse Wonder

• A video lecture or podcast longer than 15 minutes adds 1, and one longer than 30 minutes adds 2. On a paid plan the extra length is charged on top of the model, so a 45-minute lecture costs 3 on a 1× model and 4 on Claude Opus 5.5. The Create button in those creators shows the total before you start.
• The picker is the source of truth. Models are released and retired regularly, and the badge on the row you are about to choose is always current, this table is a snapshot.

Choosing a 2×, 3×, or 4× model does not draw on a separate allowance, it simply weighs more against the same one. Every paid plan can choose from the full creation model catalogue, Frontier models included.

These multipliers apply to creations only. On personal and self-serve Teams plans, AI Chat uses no AI Creation Credits: Auto and the Unlimited models are unlimited on paid plans, while Best and the Premium models draw on a separate weekly premium usage allowance. Free has three lifetime chat messages. Custom Enterprise workspaces can have separate metered terms.

A failed creation returns what it reserved, and weekly allowances reset every Monday at 00:00 UTC.

Premium usage in Chat

Chat opens on two modes:

• Auto: GPT 6 Luna, with low reasoning on Free and medium on paid plans. It is unlimited on paid plans.
• Best: GPT 6.1 Terra with medium reasoning. It uses your weekly premium usage allowance and is the paid default while that allowance remains, unless you saved a different model choice.

On desktop, Premium models contains GPT 6.1 Terra, GPT 6.1 Sol, Claude Sonnet 5.5, and Claude Opus 5.5. Unlimited models contains the other chat choices, including GPT 6 Luna, Claude Haiku 5.5, Gemini 3.8 Flash, Muse Spark 1.3, and Mistral Large 3. On phones, the picker shows the two modes; use desktop to choose a named model.

The Premium usage meter in the model menu shows the percentage left and the reset day. It measures the work performed, not a fixed number of messages. Long conversations, large sources, complex tasks, and more expensive models can use more. The usage indicators beside premium models compare their relative rate; they are not AI Creation Credit multipliers.

Premium includes a weekly allowance; Laureate includes 4× as much. Premium Lite, where available, includes a smaller allowance; Premium provides 3× as much as Lite. Allowances reset every Monday at 00:00 UTC, with no rollover. When the allowance is spent, the next premium request runs on Auto, so paid chat remains available. If you have not saved a different selection, the default returns to Best after the reset.

Background chat helpers use the same allowance when they run on a premium model. They follow the conversation's model and reasoning setting, with eligibility checked when the helper starts. Work already performed can count even if a helper is stopped, fails, or times out. Dedicated video, podcast, and other creation jobs still use their own AI Creation Credit rules. Credit packs top up creations, not premium chat usage.

Creation-only models

Some models are offered for AI creations but not for customer AI Chat. These include GPT 6 Astra, Gemini 4 Argon, Muse Wonder, Kimi K3, Grok 4.7, Gemini 3.1 Pro Preview, and Mistral Medium 3.5. Check the creation tool's picker for availability.

GPT 6.1 Sol, Claude Sonnet 5.5, and Claude Opus 5.5 are available in both places, with different usage rules: creation multipliers in creation tools, and premium usage in Chat. An old chat selection that is no longer offered falls back to your plan's default chat model without deleting the conversation.

Where you can pick a model

The Chat picker and the creation pickers are no longer the same catalogue. The chat menu is a curated selection and it is smaller than the creation catalogue, so a model you rely on for video lectures or documents may not be offered in Chat at all. Check the picker you are actually in rather than assuming a model carries across.

Model selection is available in:

• Home and AI Chat
• AI Slides and document creators
• Deep Research
• AI Video Lectures
• Podcasts
• Flashcards on paid plans
• Worksheets, Study Guides, Mind Maps, and Infographics on paid plans

Creation pickers open on three Auto rows, shown on every plan so the plan differences are visible before upgrading:

• Auto (Free): the free-tier routing.
• Auto (Premium): the routing paid plans get, Laureate included.
• Auto (Best): the highest-quality routing. It uses Claude Opus 5.5 for research, slides, video, podcasts, infographics, and flashcards, at 2 AI Creation Credits. Its credit cost follows the configured Best model.

Standard Auto uses GPT 6 Luna. Free uses low reasoning for most tools and medium for video. Paid Chat and flashcards use medium; paid video and most other creators use high. Auto (Best) is a separate 2-credit choice.

Auto rows show the tier name. Auto keeps following the current routing when defaults change. Named rows show the model name, and the credit badge remains visible before you choose.

Use best model

Creation settings carry a Use best model toggle with its credit cost shown before you turn it on, "Uses 2× AI Creation Credits", not a surprise on the bill.

More generally, whenever the model you have selected costs more than one AI Creation Credit, the creation settings show an amber warning with the exact multiple: "This uses 3× AI Creation Credits." You always see the multiplier before you press Generate.

Named models sit below Auto in creation tools. Their main menu lists Frontier Models, then Recommended Models, with the rest of the catalogue one level deeper under More Models. Chat instead groups named models by premium usage or unlimited access.

A model that needs a paid plan carries an Upgrade badge. Choosing it opens your plan options rather than failing when you send.

Scholarly remembers the model you chose for each creation tool separately, including Podcasts, Video Lectures, Research, and Slides, and remembers your Chat model separately again, so returning to a tool starts from your last pick.

In AI Chat, the Unlimited models include GPT 6 Luna, Gemini 3.8 Flash, and Claude Haiku 5.5. Luna and Haiku are available on Free; Gemini 3.8 Flash requires a paid plan. Free still has a three-message lifetime allowance regardless of the group name. The Premium models use the weekly allowance described above.

In the creation tools the Recommended Models are GPT 6.1 Sol, Grok 4.7, Claude Sonnet 5.5, Gemini 3.1 Pro Preview, Kimi K3, and Muse Spark 1.3. Flashcards recommend Gemini 3.8 Flash, Kimi K3, and Muse Spark 1.3, and also offer Claude Opus 5.5: GPT 6 Luna writes the deck, then Claude Opus 5.5 edits every card, for 2 AI Creation Credits. Free users can browse these paid choices, and selecting one opens your plan options.

Both lists change as models are released and retired, so treat the picker itself as the source of truth.

The Frontier Models group includes Claude Opus 5.5 at 2 AI Creation Credits, GPT 6 Astra at 3, and Gemini 4 Argon and Muse Wonder at 4. Any paid plan can select them in creation tools that support that model, subject to a team's model policy.

Named models

The catalogue spans the GPT, Gemini, Claude, Grok, Mistral, GLM, DeepSeek, Kimi, and Muse families, and it changes as models are released or retired, we have refreshed it several times in the last month alone. Rather than memorising a list, open the picker: every row names the model and carries its badge, so you can always see what is available to you today and what it will cost.

What each named model costs is in the table above, and on its own row in the picker.

DeepSeek V4.1 Flash is available among the named Chat models and under More Models in creation menus. It supports text and image sources and tools. It is an unlimited chat choice on paid plans and has the standard creation credit rate.

Reasoning and Thinking

Most models in Scholarly reason before they answer, and how much they reason is set for you, per feature and per plan, so there is no effort slider to get wrong. Paid plans get more deliberation than Free on the same model, and the long-form creators (video lectures, research, slides, documents) get more than the quick ones.

Two things are worth knowing about it:

• More reasoning is not automatically a better answer. It usually takes longer, and on a straightforward request it changes very little. It earns its time on hard, multi-step, or source-heavy work.
• Effort is relative to the model. How deeply one model deliberates is not directly comparable to another, so there is no single dial that makes every model "try harder".

In AI Chat, reasoning follows the selected model. There is no separate Thinking or Instant switch. Auto and Best always reason; Best uses GPT 6.1 Terra at medium effort. Where a model provides a progress summary, it collapses into a completed thought block after the answer. Old saved Thinking preferences no longer control new replies.

Chat also shows which model and reasoning setting answered a given message, so you never have to guess what produced a reply.

What the Model You Pick Actually Does

For a creation, the model you choose guides the result. On supported tools, a premium model can write the teaching plan and edit the result while GPT 6 Luna drafts it. Use best model currently selects Claude Opus 5.5 at a 2-credit base cost; it does not mean that every step runs on Opus.

Other steps in the same creation use models suited to those steps: generating images, speaking narration, transcribing audio, grading an answer. So picking Claude Opus 5.5 for a video lecture changes how the lecture is written and planned; it does not change the narrator's voice. Image generation has its own picker in Image Studio, and live Voice Mode has its own set of conversational voices.

If a provider is having a bad day, Scholarly may fall back to another model to finish your creation rather than losing it. The picker shows what you asked for, which is not a guarantee that every step and every retry used it.

Teams follow the same per-plan behaviour, within whatever model policy the workspace's Owner or Admin has set. If models are restricted or a workspace default is in place, the picker you see reflects it.

Model data handling and privacy

Private content is not used to train general-purpose AI models by default. Your material is processed by services needed to fulfil your request, subject to your plan and workspace controls. See the Privacy Policy for full details.

Tips

• If you are unsure, keep the default: Auto in creation tools, or Best in paid Chat while premium usage remains.
• Try another named model when you want a different writing style or a second take.
• Check the 2×, 3×, or 4× badge before choosing a weighted model; models without one cost 1 AI Creation Credit.
• A model missing from Chat is not a fault, check whether it is one of the creation-only models above.
• For AI creations, failed requests return their reservation, and a new flashcard deck with fewer than 15 cards does not count. Premium chat usage follows work performed and has separate rules.

Related

• AI Chat, where the chat picker lives, and what a message costs.
• Plans and Limits, AI Creation Credits and chat allowances per plan.
• Customisation Profiles, pin a preferred model to a profile so you stop re-picking it.
• Image Studio, the separate image-model catalogue, with its own picker.
• Narration Voices, who speaks a lecture or podcast, which the model choice does not decide.