The AI tab in a Bloom’s configuration controls which underlying model answers student messages. You have two options: Smart Selection, which automatically uses the latest, best-performing models we have evaluated, and Manual Selection, which pins your Bloom to a specific provider and model. Smart Selection is the right choice for almost every Bloom. Manual Selection exists for advanced users with a specific reason to deviate, like a research project, a regulatory constraint, or a subject where you have tested and prefer one model’s style.
Who can change this
The AI tab is gated by the Change Model permission. By default, this is held by the Bloom Admin roles (manager and super admin). Educators and students can open the tab, but the controls are read-only and you will see this banner:
If you need to switch a model and do not have access, ask the Space Owner or a Bloom manager to either make the change for you or grant you a higher role.
Step 1: Open the AI tab
From Manage Blooms, click into the Bloom you want to configure. In the configuration tab strip, click the AI tab (it sits between Moderation and Branding). You will see two large cards stacked vertically: Smart Selection on top, marked with a green Recommended pill, and Manual Selection below.
Step 2: Use Smart Selection (recommended)
Smart Selection is the default for new Blooms. When it is active, Bloom routes chat messages through the strongest model we have validated for general teaching tasks. The card explicitly tells you which models are in rotation:
The benefits listed on the card:
- Auto-updated to newest models: when we promote a new frontier model, every Bloom on Smart Selection moves with us. You do not need to revisit this page.
- Optimized for safety and performance: the rotation is paired with our content moderation stack and tuned for educational dialogue.
If you change nothing else on this tab, you will get a good experience. Most Blooms (including the ones on our paid pilot programs) leave this on.
Step 3: Switch to Manual Selection
Click the Manual Selection card to expand it. Two dropdowns appear: AI Provider and AI Model. Once you switch off Smart Selection, your Bloom will not change models until you come back here and update them.
AI Provider
The provider list is grouped by the underlying API:
- Azure: Microsoft’s Azure OpenAI deployments. This is the default for new manual configurations and offers the broadest model range.
- OpenAI: direct OpenAI API. Same model family as Azure but routed through OpenAI’s own endpoints.
- Google: Gemini models, including 3.1 Pro Preview, 3 Flash, and the 2.5 family.
Switching providers resets the model dropdown, so pick a provider first, then a model.
AI Model
Each provider exposes a curated list. For Azure the chat-eligible options include GPT-5.2 (the default), GPT-5.1, and GPT-5. For Google, the headline options are Gemini 3.1 Pro Preview, Gemini 3 Flash, Gemini 2.5 Pro, and Gemini 2.5 Flash. Models that we have deprecated for chat (embedding models, very small classifiers) are filtered out of the menu, so the list you see is the list you can safely pick.
Step 4: Choosing a model
There is no universally best model. A few rules of thumb that hold up in practice:
- Default to GPT-5.2 (Azure) for general subjects. It handles long context, follows instructions tightly, and is the default chat model behind Smart Selection.
- Try Gemini 3.1 Pro Preview for STEM Blooms with heavy mathematical or code-heavy content. It is also the model Deep Think uses on Google when reasoning mode is invoked.
- Try Gemini 3 Flash when latency matters more than depth, for example a busy intro course where students want quick clarifications.
- Avoid older or experimental tagged models for live student traffic. They are kept available for evaluation and migration but are not what we recommend for production.
Click Save at the bottom of the configuration page once you are happy. The change applies to the next message. Conversations already in progress continue on whatever model their in-flight request was already using.
Common issues
The whole tab is locked with a Lock icon
You do not have the Change Model permission. This is expected for educators and students. Ask a manager or super admin to make the change.
The Save button does nothing or shows a validation error
If you toggled to Manual Selection but never picked a provider or model, the form will fail validation. Either pick a provider and model, or switch back to Smart Selection.
Manual Selection picked, but answers feel different to what I tested
Two things to check. First, Deep Think (when enabled by the student or educator on a message) swaps the chat model for a stronger reasoning model for that message only. Second, certain tasks like intent classification and title generation always use small purpose-built models, regardless of your chat model choice. Your chat model only governs the main conversational reply.
I want a model that is not in the dropdown
The list reflects what we have validated, deployed, and deemed safe for student traffic. If a model you want is missing, send us a request through the in-app feedback. We add new models to the rotation on a regular cadence.
What’s next
- Understand how the per-message reasoning model works → How Deep Think works
- See the full configuration tour → Configuration overview
- Open Bloom and tune a model now → Manage Blooms