Supported Models

The following list of models details models that have been tested for compatibility with DataPrompt.

ModelDeveloperAvailable viaAccessCapabilities

Model names are the current and recent releases from each developer. Availability follows the provider and the curated catalogue; the model picker in the app is the authoritative list for your workspace. Codex subscription models are described on theintegrations page.

Providers

Connecting a provider

Open Connections, add a model provider connection, paste the key, and verify it. The models you can select are the provider's live catalogue filtered by DataPrompt's curated list.

ProviderAccessConnectNotes
OpenAIOwn keyResponses APIAdd an OpenAI API key under Connections and verify it.Batch mode applies to stateless assistant agents only. Embeddings for knowledge sources also run here.
Azure OpenAIManagedResponses APIConfigured by DataPrompt for managed and enterprise deployments against your Azure region.Deployment is enabled per environment once the Azure deployment is confirmed.
AnthropicOwn key: in qualificationMessages APIAvailable from the curated catalogue today. Own-key connections are accepted but wait on qualification of the Messages protocol.Own-key runs fail closed until the protocol is qualified, rather than silently falling back.
Google GeminiOwn keyChat with toolsAdd a Gemini API key. Own-key requests use Google's OpenAI-compatible endpoint.The native Interactions protocol is used by the curated catalogue; own keys use the chat-with-tools protocol.
OpenRouterOwn keyChat with toolsAdd an OpenRouter key to reach open-weight and third-party models through one account.Streaming to the workspace is not supported on OpenRouter. Each agent profile pins one physical model.
NVIDIA NIMOwn keyChat with toolsAdd an NVIDIA API key for hosted open-weight models.The default model for new workspaces runs here.
DeepSeekOwn keyChat with toolsAdd a DeepSeek API key.64k context.
Alibaba QwenOwn keyChat with toolsAdd a DashScope API key. The international endpoint is used.128k context.