AI Models & Defaults
See which model runs your chat or AI creation, and how much reasoning it uses. These settings cover AI Chat and the AI creation agents: flashcards, video lectures, podcasts, research, slides, documents, infographics, and mind maps.
Open a feature's model selector to see its current choices. Each model shows its reasoning effort beneath its name. Auto also names the model it currently uses. Your saved model choice or a customization profile can replace the default.
What reasoning effort means
Reasoning effort is the level of deliberation requested from the model. Low, medium, high, and extra-high are relative settings within that model, not seconds of thinking or a quality score. Low on one model is not directly comparable to low on another. Higher effort can take longer; it does not guarantee a better answer.
Thinking off means Chat uses the model's non-thinking mode. Reasoning not available means that model has no Thinking mode. Provider default means Scholarly does not specify an effort level.
Chat defaults
| Plan | Auto model | Starting setting | With Thinking on |
|---|---|---|---|
| Free | Claude Haiku 5.5 | Thinking off | Optional Thinking requires a paid plan |
| Premium | Claude Haiku 5.5 | Thinking off | Low |
| Laureate | Claude Haiku 5.5 | Thinking off | Low |
Chat remembers your Thinking preference. On a paid plan, turn Thinking on or off in the model selector for models that support both modes. Models with reasoning always on keep it on, including on Free. The effort label for an optional-thinking model says “with Thinking on”; it does not mean Thinking is currently enabled.
AI creation defaults
Each cell shows model · reasoning effort. These are the plan's Auto settings when that feature is available; this table does not grant access to features outside your plan.
| Feature | Free | Premium | Laureate |
|---|---|---|---|
| Flashcards | GPT 6 Luna · Low | GPT 6 Luna · Medium | GPT 6 Luna · Medium |
| Video Lectures | GPT 6 Luna · Medium | GPT 6 Luna · High | GPT 6 Luna · High |
| Podcasts | GPT 6 Luna · Low | GPT 6 Luna · High | GPT 6 Luna · High |
| Deep Research | GPT 6 Luna · Low | GPT 6 Luna · High | GPT 6 Luna · High |
| Slides & document agents | GPT 6 Luna · Low | GPT 6 Luna · High | GPT 6 Luna · High |
| Infographics & Mind Maps | GPT 6 Luna · Low | GPT 6 Luna · High | GPT 6 Luna · High |
Document agents include PDF Documents, Study Guides, Worksheets, Timelines, SOPs, Flowcharts, Lesson Plans, Outlines, Cheat Sheets, Reports, Case Studies, Rubrics, Spreadsheets, File Documents, Essay Writer, and Story Books. Creating a source deck for Practice Quiz follows the flashcard defaults; question generation and grading are separate steps.
Auto (Best), also available through Use best model, uses Claude Sonnet 5.5 · Medium for research, AI Slides, video lectures, podcasts, and infographics/mind maps. It costs 3 AI Creation Credits per creation. Standard Auto uses 1 credit. On a paid plan, a video lecture or podcast longer than 15 minutes adds 1 credit to either. Document creators other than AI Slides currently use the plan default when Sonnet or Best is selected.
Premium Teams and Laureate Teams follow their corresponding plan settings, subject to the workspace's model policy. If an administrator restricts models or sets a workspace default, the workspace's selector reflects that policy.
Reasoning for each named model
These are the configured efforts when you pick the model by name. — means the model is not offered on that surface. “Other agents” covers research, slides/documents, video, infographics/mind maps, and podcasts, where that model is offered; their catalogs differ.
| Named model | Chat | Flashcards | Other agents |
|---|---|---|---|
| GPT 6 Luna | Low on Free; Medium on paid plans, always on | Low on Free; Medium on paid plans | Low on Free (Medium for video); High on paid plans |
| Claude Haiku 5.5 | Low with Thinking on | — | — |
| Claude Sonnet 5.5 | — | — | Medium |
| Claude Opus 5.5 | — | — | Low |
| GPT 6.1 Sol | — | — | Low |
| GPT 6 Astra | — | — | Low |
| Gemini 3.8 Flash | Low with Thinking on | Medium | Low |
| Gemini 3.5 Flash Lite | Low, always on | — | Low |
| Gemini 3 Flash | Low, always on | Low | Low |
| Gemini 2.5 Flash | — | Low | Low |
| Gemini 3.1 Pro Preview | — | — | Low |
| Grok 4.7 | — | — | Low |
| GLM 5.3 Flash | Low, always on | Low | Low |
| DeepSeek V4.1 Flash | Low with Thinking on | Low | Low |
| Kimi K3 | — | Low | Low |
| Muse Spark 1.3 | Low, always on | Low | Low |
| Mistral Small 4 | Low with Thinking on | — | — |
| Mistral Medium 3.5 | — | — | Low |
| Mistral Large 3 | Reasoning not available | — | — |
The effort is set for the feature and plan; selecting a named model does not add a separate effort slider. In Chat, the Thinking toggle changes whether supported optional reasoning is used.
What the selected model covers
For a creation, this is the main agent that writes or directs the result. Tasks such as producing images, generating speech, transcribing sources, or grading answers can use separate models. A video or podcast model choice therefore does not name its speech model. This guide covers the main agents and text Chat, rather than the separate Image Playground or live Voice Mode catalogs.
If a provider fails, Scholarly may retry with a fallback model. The selector shows the requested model and effort, not a guarantee that every step or recovery attempt uses that model. Defaults and available choices can change; the selector receives the current catalog for your account.
Related
- Choosing an AI Model — model access, credit multipliers, and saved choices.
- AI Chat — chat, sources, and Thinking.
- Customization Profiles — reuse a preferred model.