Provider and Model Support
How each AI framework abstracts over different model providers.
Vercel AI SDK
Ships roughly 38 first-party, individually typed provider packages, covering most major model vendors with dedicated, purpose-built integrations.
TanStack AI
Ships a smaller set of official provider adapters (around nine, including OpenAI, Anthropic, Gemini, Ollama, Groq, and xAI), reaching much broader model coverage through OpenRouter and a generic OpenAI-compatible adapter for any compatible endpoint.
LangChain
Supports an enormous number of providers and integrations through its community-contributed integration ecosystem, arguably the broadest of any framework in this category.
LangGraph
Inherits LangChain’s model and provider integrations directly, since it’s built on top of LangChain’s chat model abstractions rather than defining its own.
LlamaIndex
Supports a wide range of LLM and embedding providers through its own integrations layer, with particular depth around embedding models and vector-store providers for retrieval.
Mastra
Supports the major providers (OpenAI, Anthropic, Google, and others) largely through the Vercel AI SDK’s model providers under the hood rather than maintaining a fully separate provider layer.
CrewAI
Supports most major LLM providers, commonly used with OpenAI, Anthropic, and local models served through Ollama.
Semantic Kernel (Microsoft Agent Framework)
Supports OpenAI, Azure OpenAI, and other providers through a connector model, with particularly deep first-party support for Azure AI services.
Haystack
Supports major LLM providers plus a wide range of document stores and retrieval backends (Elasticsearch, OpenSearch, Weaviate, and others) as pluggable pipeline components.