Capability matrix

This matrix describes what the current Anvia provider adapters can represent. A Yes means the adapter implements the Core capability; it does not mean every upstream model ID, account, region, or endpoint supports it.

Always test the exact production model and configuration, including provider-specific additionalParams.

Completion capabilities

Capability	OpenAI	Anthropic	Gemini	Mistral	Grok
Text completion	Yes	Yes	Yes	Yes	Yes
Streaming	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
Image input	Yes	Yes	Yes	No	Yes
Document-file input	Responses only	Yes	Yes	No	Responses only
Reasoning content	Yes	Yes	Yes	No	Yes
Provider-executed tools	Responses only	No	No	No	Responses only
Normalized citations and provider-tool events	Responses only	No	No	No	Responses only

OpenAI and Grok expose Responses and Chat completion adapters. Responses is the default for their normal first-party clients. Their Chat adapters do not declare document-file input or provider-executed tools.

Document-file input means a provider-native file or URL attachment. Text extracted from a document can still be included as normal text with any completion provider.

Other model 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	Yes
Transcription	Yes	No	Yes	No	Yes
OCR	No	No	No	Yes	No
Model listing	Yes	Yes*	Yes	Yes	Yes

* Direct AnthropicClient can list models. AnthropicVertexClient cannot because Vertex AI does not expose Anthropic's Models API.

These capabilities use separate Core interfaces and provider factories. Support for completion does not imply support for embeddings, transcription, or media generation.

Read a completion model declaration

Anvia completion models expose adapter capabilities on the model object:

const model = client.completionModel(config.model)

console.log(model.provider)
console.log(model.defaultModel)
console.log(model.capabilities)

A typical guard can reject an invalid application configuration at startup:

if (config.attachments && !model.capabilities.documentInput) {
  throw new Error('Configured completion model does not accept document files')
}

if (config.structuredOutput && !model.capabilities.outputSchema) {
  throw new Error('Configured completion model does not expose output schemas')
}

This is contract validation, not a network capability probe. The provider may still reject an unsupported model, account, region, parameter, file type, or feature combination.

Compatible endpoints

Compatible APIs share an OpenAI or Anthropic HTTP shape, but that does not guarantee feature parity. Streaming chunks, tool arguments, tool choice, structured output, reasoning fields, and media inputs commonly differ.

When OpenAIClient receives a custom baseUrl, it defaults to the Chat adapter. Set completionApi explicitly only when the endpoint actually implements the selected surface.

Minimum verification

After using this table to make a shortlist, run the smallest request that proves every required path:

• a direct completion for credentials and request mapping
• a streamed completion consumed through the final event
• a forced or required tool call when the workflow depends on tool choice
• a parsed completion for structured output
• representative image, document, audio, embedding, transcription, or OCR input
• an error case to confirm retry and fallback behavior

Record the provider, exact model ID, endpoint, region, and relevant parameters with the test result. Repeat these tests when any of them changes.