Managing Models
On the page Management Environment → AI Settings → Models, you decide which AI providers are available in the environment. Users then choose a preferred provider themselves under Preferences. The concrete models come from AI-School's central model catalog and are no longer manually chosen per environment from a models list.
The models only work if the corresponding API keys are properly configured for the environment.
Background tasks
At the top of the page you configure which provider is used for general background tasks. These are processes the application performs without a user actively selecting a model.
| Task | Example |
|---|---|
| Summaries | Automatic summaries of documents and chats |
| Text from images | Extract text and information from images or PDF pages |
| Audio transcription | Convert spoken audio to text |
| Embeddings | Vector representations for search functionality |
For background tasks, the environment can choose between OpenAI and European AI. When changing, the application shows a warning first, as this affects new system processes. Existing embeddings are not automatically regenerated after switching providers.
Provider preferences
The main setting consists of provider cards. Each card represents a provider or a combination of providers:
| Provider card | Used for |
|---|---|
| OpenAI | OpenAI language, image, and speech models |
| Claude | Claude language models, combined with Google image and speech models |
| Gemini language models, Google image models such as Nano Banana and Nano Banana 2, and Google TTS | |
| European AI | Mistral language and speech models and Black Forest Labs/FLUX image models such as FLUX.2 Flex and FLUX.2 Pro |
Each card includes a short explanation of the provider. You also see which language, image, and speech models are available with the chosen setting. Click a model button to view more information about that model.
Current model catalog
Language models
| Provider card | Cost-effective | High quality | Optional top model |
|---|---|---|---|
| OpenAI | GPT-5.6 Luna | GPT-5.6 Terra | GPT-6 Astra |
| Claude | Claude Haiku 4.5 | Claude Sonnet 5 | Claude Fable 5.1 |
| Gemini 3.5 Flash-Lite | Gemini 3.8 Flash | Gemini 3.1 Pro Preview | |
| European AI | Mistral Small 4 | Mistral Medium 3.5 | Mistral Large 3 |
Image and speech models
| Provider card | Cost-effective image | High quality image | Speech model |
|---|---|---|---|
| OpenAI | GPT Image 2 | GPT Image 2.5 Flare | GPT-4o mini TTS |
| Claude | Nano Banana 2 Lite | Nano Banana 2 | Gemini 3.1 Flash TTS Preview |
| Nano Banana 2 Lite | Nano Banana 2 | Gemini 3.1 Flash TTS Preview | |
| European AI | FLUX.2 Flex | FLUX.2 Pro Preview | Voxtral Mini TTS |
Quality per provider
For each provider choose one of the following settings:
| Setting | Meaning |
|---|---|
| High quality | The provider's stronger general model is available. This is suitable for more complex questions, analysis, and careful answers, but may be more expensive. |
| Cost-effective | The provider's smaller or cheaper model is available. This is often faster and cheaper but may perform less well on tricky questions. |
| Combination | The cost-effective and higher-quality model are available. The cost-effective model is shown first and is the default choice. Optionally you can also add the top model. |
| Disable | The provider is completely unavailable to users. The associated models also do not appear in Preferences or model choices. |
If you change a provider setting, the application will first ask for confirmation. The confirmation briefly explains the impact on quality, cost, and availability.
Add top models
Under Combination you can additionally mark that the provider's top model becomes available. This option is off by default. Enable this only when users should have conscious access to the highest quality and the higher cost is acceptable.
The top models are GPT-6 Astra for OpenAI, Claude Fable 5.1 for Claude, Gemini 3.1 Pro Preview for Google, and Mistral Large 3 for European AI. This option only adds an extra language model; image and speech models remain linked to the chosen provider card.
Image and speech models
The central model catalog includes, in addition to language models, also models for:
- Text to image: including supported format choices Auto, Square, Landscape, and Portrait.
- Text to audio: including available voices and language support.
Users see these models in their preferences and in the model chooser on the dashboard. The provider they choose as a preference determines which linked language, image, and speech models are used by default.
Order of providers
With the arrow buttons on the provider cards you set the preferred order. This order is used when assembling the available models in the environment.
If no settings have been saved yet, AI-School uses this default order:
- OpenAI
- Claude
- European AI
Save
Changes are saved with the Save button in the app toolbar. After saving, the available models are re-determined from the central catalog and the environment's provider settings.
The environment no longer manages separate model records for users. AI-School keeps the central model catalog up to date. This means model names and identifiers can be updated without users having to re-select a specific model.
Help with setup
AI-School can assist in choosing a suitable provider strategy. For many environments, High quality is a safe default. Choose Cost-effective when cost and speed are more important. Choose Combination when users should generally work cheaply, but can switch to a stronger model for more complex tasks. Enable the top model within Combination only for environments where maximum quality is more important than cost control.
AI usage and prices
Central administrators can use AI usage to view measured server usage by month and environment and manage model prices with an effective date. Costs are calculated afterwards. A missing price does not mean zero usage. Existing invoicing remains unchanged.
The monthly overview shows one total per customer environment and a grand total across all filtered environments. Models are combined; different currencies and usage units remain separate.
Click a customer environment to view monthly usage by AI model in a dialog. The customer total includes all model pages.
In the customer dialog, switch between totals by school and by AI model. Click a school to see its model totals. New measurements retain the request school or, if absent, an unambiguous school from the user profile. Measurements without a school stay separate; existing test measurements are not backfilled.
Model prices also show cache rates, separate image token rates and higher rates above a context threshold. These optional fields can be edited in the price dialog. The source link points to the provider. Missing usage details leave the calculated cost unavailable.
Separate model settings for students
Students inherit the general model settings. Choose different models here to manage usage.
