Skip to main content

Supported providers

DIAL ships three model adapters for provider connectivity, API translation, and supported passthrough paths, all served by one service, DIAL Model Provider. This page lists the providers each adapter supports, the model families available through each, and the authentication methods required.

Adapter-provider mapping​

ProviderAdapter in DIAL Model ProviderAuthentication
OpenAI PlatformOpenAI Adapter, /compat/openaiAPI key
Azure AI FoundryOpenAI Adapter, /compat/openaiAPI key or Azure AD token
MistralOpenAI Adapter, /compat/openaiAPI key
Anthropic APIOpenAI Adapter, /compat/openaiAPI key
DatabricksOpenAI Adapter, /compat/openaiAPI key
Fireworks AIOpenAI Adapter, /compat/openaiAPI key
Alibaba Cloud Model StudioOpenAI Adapter, /compat/openaiAPI key
vLLM / OllamaOpenAI Adapter, /compat/openaiNone (local)
AWS BedrockBedrock Adapter, /compat/bedrockAWS IAM credentials
Google Vertex AIVertex AI Adapter, /compat/vertexaiGCP service account or ADC

Models by adapter​

OpenAI Adapter​

The OpenAI Adapter (v0.43.3) selects behaviour from the upstream API a deployment points at, not from the model's name. So it is listed here by upstream API rather than by model: any deployment that matches none of the specialised categories below is treated as OpenAI-compatible text-to-text chat completion or text embeddings.

Upstream APIReachable through
Chat CompletionsAzure OpenAI, OpenAI Platform, Azure AI Foundry, Mistral, Databricks, vLLM, Alibaba Cloud Model Studio, Amazon Bedrock
ResponsesCurrently Azure OpenAI only; see API support
Anthropic MessagesClaude on Azure AI Foundry, the Anthropic API, and other providers of the same API
EmbeddingsAzure OpenAI, OpenAI Platform, vLLM, Azure multimodal embeddings
Image generationdall-e-3, gpt-image-1
Videosora, sora-2
Text to speechtts, tts-hd, gpt-4o-mini-tts
Speech to textwhisper, gpt-4o-transcribe, gpt-4o-mini-transcribe, gpt-4o-transcribe-diarize, Qwen3-ASR on vLLM

GPT-4 Vision is deprecated — those models were retired by both Azure OpenAI and the OpenAI Platform.

Bedrock Adapter​

The Bedrock Adapter (v0.43.3) supports AWS Bedrock through both the Runtime API and Converse API.

VendorModel families
AnthropicClaude 5 (Opus, Sonnet, Fable), Claude 4.8 Opus, Claude 4.7 Opus, Claude 4.6 (Opus, Sonnet), Claude 4.5 (Sonnet, Haiku), Claude 4.1 Opus, Claude 4 (Opus, Sonnet), Claude 3.7 Sonnet, Claude 3.5 (Sonnet, Haiku), Claude 3 (Opus, Sonnet, Haiku)
MetaLlama 4 (Scout, Maverick), Llama 3.3, Llama 3.2, Llama 3.1, Llama 3
DeepSeekR1
MiniMaxM2.5
Stability AISD 3.5 Large, Stable Image Ultra/Core
AmazonNova Pro, Nova Lite, Nova Micro
AI21 LabsJamba 1.5 (Large, Mini)
CohereCommand R, Command R+
EmbeddingsTitan Multimodal G1, Titan Text V2, Titan Embeddings G1, Cohere Embed (English, Multilingual)

Each chat model is also addressable through cross-region inference by prefixing its deployment id with a region — for example us.anthropic.claude-sonnet-4-5-20250929-v1:0.

Vertex AI Adapter​

The Vertex AI Adapter (v0.39.1) supports Google Vertex AI, Google AI Platform, and Google Search grounding for Gemini models.

CategoryModel families
Gemini3.5, 3.1, 3, 2.5, 2.0
ImagenImagen 4, Imagen 3, Imagen 2
VeoVeo 3.1, Veo 3.0
Partner modelsAnthropic Claude, Mistral Medium 3, Mistral Small, Codestral
Embeddingsgemini-embedding-001, text-embedding-004/005, text-multilingual-embedding-002, multimodalembedding@001

OpenAI-compatible models​

Any model that exposes an OpenAI-compatible API can be accessed through the OpenAI Adapter. This includes self-hosted models served by vLLM, Ollama, TGI, and similar inference servers. Configure the adapter's endpoint to point to the model server and use the appropriate deployment environment variables.

For adapter environment variable reference, see DIAL Model Provider configuration.

How to enable a model​

Models are registered as deployments in DIAL Core through one of two methods:

  • DIAL Core config.json — add a model entry whose baseUrl points at the adapter's DIAL Model Provider path, with its upstream configuration. See the config.json models reference.
  • DIAL Admin — add models through the administration UI. See Admin Panel: Models.

Adding an unsupported provider​

Organizations that need to integrate a provider not covered by the official adapters can build a custom adapter using DIAL SDK. The Tutorial: custom adapter walks through the process from project setup to deployment.

Further reading​