Providers
Capability matrix
Compare the current Anvia provider adapters by exposed runtime capability.
This matrix describes what each Anvia provider adapter exposes today. It is adapter-level truth, not a guarantee that every upstream model id supports every capability.
Always smoke test the exact provider model id, account, region, endpoint, and additionalParams your workflow will use.
Completion Capabilities
| Capability | OpenAI | Anthropic | Gemini | Mistral | Grok |
|---|---|---|---|---|---|
| Text completion | yes | yes | yes | yes | yes |
| Streaming completion | yes | yes | yes | yes | yes |
| Tools | yes | yes | yes | yes | yes |
| Tool choice | yes | yes | yes | yes | yes |
| Structured output schema | yes | no | yes | yes | yes |
| Chat image input | yes | yes | yes | no | yes |
| Chat document file input | Responses yes, Chat adapter no | yes | yes | no | Responses yes, Chat adapter no |
| Reasoning content | yes | yes | yes | no | yes |
OpenAI has two completion adapters. OpenAIResponsesCompletionModel is the default for normal OpenAI API usage. OpenAIChatCompletionModel is available when a workflow needs the chat-completions adapter.
Grok also exposes Responses and Chat adapters; Responses is the default.
Text document content and static context can still be represented as text. The document row above is about file or URL document attachments that require provider-native document input support.
Non-Completion Capabilities
| Capability | OpenAI | Anthropic | Gemini | Mistral | Grok |
|---|---|---|---|---|---|
| Embeddings | yes | no | yes | yes | no |
| Image generation | yes | no | yes | no | yes |
| Audio generation | yes | no | no | no | no |
| Transcription | yes | no | yes | no | no |
| OCR | no | no | no | yes | no |
| Model listing | yes | yes | yes | yes | yes |
Provider-specific model factories return objects that implement core contracts such as EmbeddingModel, ImageGenerationModel, AudioGenerationModel, TranscriptionModel, or ModelListingClient.
Compatible API Guides
| Guide | Endpoint option | Notes |
|---|---|---|
| OpenAI-Compatible | baseUrl |
Defaults to the chat-completions adapter when set. Use completionApi to force "responses" or "chat". |
| Anthropic-Compatible | baseUrl |
Use for Anthropic-compatible APIs. |
Compatible APIs need extra capability checks. A compatible API can share an HTTP shape while differing in tools, tool choice, schemas, reasoning fields, streaming chunks, or media support.
Practical Selection
Use the matrix to choose candidates, then prove the workflow with tests:
- direct completion for credentials and basic request mapping
- streaming completion if UI progress matters
- forced or required tool calls if tools are enabled
- parsed completion if structured output is required
- sample image, audio, transcription, embedding, or OCR requests for media workflows
- model listing only for inventory, not capability proof
Read Testing strategy for provider smoke tests and Model listing for inventory behavior.