Models and response settings
Each Q&A and Chat slide answers with an AI model, tuned by a few response settings. You set workspace defaults once, and each slide can follow them or use its own values. This page is for authors who configure slides and for owners who set the defaults.
Where you choose a model
Section titled “Where you choose a model”| Level | Where | Who can change it |
|---|---|---|
| Workspace defaults | Settings › AI & models › Defaults for new slides | Owner. Editors see the values read-only. |
| One slide | Slide editor, Which AI answers? section, Adjust | Owners and Editors. |
In the slide editor, the Which AI answers? section sums up the current choice: the model, then SlideMind Credits or “via” the provider, and a line such as “balanced, up to 1024 tokens”. Select Adjust to open the AI model panel and change the settings described below.
Capture slides never call the AI, so they have no model.
The model catalogue
Section titled “The model catalogue”The models you can choose depend on how the workspace reaches AI. See Choosing how AI is funded.
In SlideMind Credits mode
Section titled “In SlideMind Credits mode”You choose from a curated catalogue maintained by SlideMind. It changes over time, so this documentation does not list model names. The catalogue has two groups:
- Fast & economical: quick, inexpensive models, well suited to most quiz feedback and short conversations.
- Most capable: stronger models for complex coaching or feedback, at a higher cost.
Each model shows its relative speed (fast, balanced speed or slower) and cost (low cost, medium cost or higher cost). A model marked Included with paid plans cannot be selected on your current plan.
When SlideMind withdraws a model from the catalogue, slides that still use it keep working: the picker shows the model as unavailable with “No longer offered”, and the catalogue’s default model answers until you choose another one.
With your own API keys
Section titled “With your own API keys”The list comes from your provider’s own catalogue, once its key is Valid. Until then, the Default model picker in Settings › AI & models says “The [provider] key must pass validation before its model catalogue can be loaded.” See Your own provider keys.
Type in the picker to search by model name, ID or provider. If a saved model is no longer offered by the provider, the picker shows it as Unavailable with “This saved model is no longer available and will not be replaced automatically.” Choose another model and save.
Slide settings
Section titled “Slide settings”The AI model panel of the slide editor contains these settings:
| Setting | What it does | Example |
|---|---|---|
| Provider | Only with your own API keys. Inherited follows the workspace’s default provider; you can also pick any provider that has a valid workspace key. | Inherited (OpenRouter) |
| Model | The model that answers. Workspace default (…) follows the workspace default model. | A model from the Fast & economical group |
| Reasoning effort | How much the model thinks before it answers, for models that support it. Inherit workspace default follows the workspace value. | Low for quick quiz feedback |
| Temperature | How varied the answers are, from 0 to 2. | 0.3 for consistent corrections |
| Max tokens | The maximum length of one AI reply. | 400 for a short feedback paragraph |
A slide that follows the workspace default model or reasoning effort changes when the owner changes the default. A slide with its own choice keeps it.
Reasoning effort
Section titled “Reasoning effort”The options depend on the model. The full range is Automatic, Off, Minimal, Low, Medium, High, Extra high and Maximum; each model offers only the levels it supports, and some models cannot turn reasoning off.
- Automatic depends on how the workspace pays for AI. With SlideMind Credits, it means a low effort on the models that support it, so replies come faster and cost less. With your own provider keys, it leaves the level to the model’s standard setting. Choose Medium or High explicitly when a slide needs more reasoning.
- “Higher reasoning effort generally increases latency and cost.” For short feedback, a low level is usually enough.
Temperature
Section titled “Temperature”The slider shows the value. The summary of the Which AI answers? section names the matching style, for example “balanced, up to 1024 tokens”:
| Value | Style | Use it for |
|---|---|---|
| 0 to 0.3 | focused | Quiz corrections, factual answers, consistent feedback |
| Above 0.3, up to 0.9 | balanced | Most explanations and coaching |
| Above 0.9 | creative | Open role-plays and varied conversations |
Some models do not accept a temperature when they reason. The setting is then disabled with the note “Temperature is unavailable for this model and reasoning combination.”
Max tokens
Section titled “Max tokens”Max tokens caps the length of each reply, between 1 and 32,768. More tokens allow longer feedback but can increase cost; too few may cut an answer short. A token is roughly a short word or part of a word.
Cost estimate
Section titled “Cost estimate”Under the settings, the slide editor estimates the cost of the slide with its current instructions, model and max tokens:
- for a Q&A slide, ”~… per exchange”;
- for a Chat slide, “Up to ~… per conversation”, using the slide’s maximum number of turns.
The estimate shows credits in SlideMind Credits mode and USD with your own keys. It is an estimate only: real conversations decide the actual cost. When the model has no public pricing, the slide editor says so and shows no estimate.
Workspace defaults
Section titled “Workspace defaults”In Settings › AI & models › Defaults for new slides, the owner sets:
- Default model: the model of every slide set to Workspace default (…). With your own keys, the list comes from the default provider. On SlideMind Credits, the catalogue’s default model is preselected until you choose another one.
- Reasoning effort: the effort of every slide set to Inherit workspace default.
- Temperature and Max tokens: default response values.
Save settings stays inactive until you change a value. Select it to apply. Slides that follow the defaults use the new model and reasoning effort right away, including slides already published. If you leave the tab without saving, SlideMind asks Leave with unsaved changes?: Discard changes drops them.
Plain-text AI responses
Section titled “Plain-text AI responses”Authoring tools display text as-is, so Markdown formatting such as **bold** would show its symbols to learners. In Settings › AI & models › Plain-text AI responses, two switches remove Markdown formatting (headings, bold, lists, links, code) from AI responses before they reach authoring tool variables and the SlideMind widget:
- Strip Markdown in Q&A mode
- Strip Markdown in Chat mode
Both are on by default. Structured JSON outputs are never affected. Only the owner can change them. They belong to the same form as the defaults for new slides: select Save settings to apply.
Next steps
Section titled “Next steps”- Write the AI instructions: get better answers from any model.
- Try it: test a slide’s settings before you publish.
- SlideMind Credits: how model choice affects credit use.
- Your own provider keys: connect a provider and load its catalogue.