You can leverage Glean’s generative AI features using either a Glean key - managed access to large language models (LLMs) for Glean Assistant and Agents. Or, you can leverage your own model deployments with the customer key option. We recommend that customers use the Glean key as it allows for secure access to large language models while reducing the burden of managing machine learning models directly. Key benefits include:
  • Consistent Performance: The Glean key enables customers to achieve dependable AI performance without needing to manage the underlying models, as it utilizes a combination of dedicated and on-demand capacities. Additionally, Glean determines the best model that should be used for Glean Assistant based on thorough evaluation of the model families.
  • Automatic Model Upgrades: Glean handles updates and improvements to models, ensuring users have access to the latest advancements without additional effort.
  • Zero data retention: Inputs and outputs related to model usage are not logged, and customer data is not used for training, enhancing privacy and trust.
When you choose the customer key option, you are responsible for model configuration, selection upgrades, and capacity planning which can result in increased support complexity. We recommend provisioning model capacity to avoid any disruption to the Glean service. Customer Key can be set up in Admin Console > Platform > LLM Glean currently supports the following LLMs for Assistant:

Provider

Large Model

Small Model

Azure OpenAI or OpenAI Direct

GPT 4.1 (Recommended)

GPT 4o-mini (Recommended) or GPT 4.1-mini

GPT 4o

GPT 4o-mini

Vertex AI

Claude Sonnet 4 (Recommended)

Claude Sonnet 4 (Recommended)

Claude Sonnet 3.7

Claude Haiku 3.5

Claude 3.5 Sonnet V2

Claude 3.5 Haiku

Gemini 2.5 Pro

Gemini 2.5 Flash

Gemini 2.0 Flash

Gemini 2.0 Flash

Bedrock

Claude Sonnet 4 (Recommended)

Claude Sonnet 4 (Recommended)

Claude Sonnet 3.7

Claude Haiku 3.5

Claude 3.5 Sonnet V2

Claude 3.5 Haiku

Looking for the original version of this page? You can find the archived version here.