LLM requirements for Context Grounding

This page covers LLM provider compatibility, supported model versions, and capability requirements for each Context Grounding feature.

For the Bring your own model (BYOM) setup, refer to Configuring LLMs for your product.

Provider support

The following table shows which providers are supported for each Context Grounding feature.

Feature	OpenAI	Google Gemini	Anthropic Claude	Self-hosted OSS
Advanced Extraction	Supported	Supported	Supported	Not supported
Batch Transform	Supported	Supported	Supported	Supported
Batch Transform with Web Search	Not supported	Supported	Not supported	Not supported
DeepRAG	Supported	Supported	Supported	Not supported
Embeddings	Supported	Supported	Supported	Supported

OpenAI V1 compliant model requirements

When using a custom model via the OpenAI V1 Compliant LLM connector, the model must meet the following requirements for each Context Grounding feature.

Feature	Requirements	Recommended
Embeddings	< 4,096 dimensions; ≥ 8k input tokens	gemini-embedding-001
Advanced Extraction	Must have: Forced tool calling; multimodal image support. Recommended: 16k input tokens; 32k output tokens	gemini-2.5-flash
DeepRAG	Must have: Forced tool calling; count > 1 support. Recommended: 1M input tokens; 64k output tokens	gemini-2.5-flash
Batch Transform	Must have: Forced tool calling. Recommended: 128k input tokens	gemini-2.5-flash (smart), gemini-2.5-flash-lite (fast)
Batch Transform with Web Search	Gemini web search tool	Any Gemini web search model

Must have — without this capability, the feature does not function.
Recommended — without this capability, the feature is untested and may perform worse or fail more often.