# Supported LLM Models

The AI Agent Foundry supports the following LLM Models:

**Inference models**

* Anthropic Claude family
  * Claude Sonnet 4.6 (except Korea region)
  * Claude Sonnet 4.5 (except Korea region)
  * Claude Haiku 4.5 (except Korea region)
  * Claude Sonnet 4
* OpenAI GPT family
  * GPT-5.5
  * GPT-5
  * GPT-5 mini
  * GPT-5 nano
  * GPT 4.1


The OpenAI GPT family is not enabled by default. If you wish to use it, please have an Owner or Administrator request activation from Technical Support or your Customer Success Manager.

**Image generation models**

* Amazon Nova
  * Nova Canvas


End of Life
Nova Canvas is being retired by AWS and is reaching end of life in AI Agent Foundry no later than the end of September 2026 — possibly sooner, since AWS automatically disables a Bedrock model after 30 days without use and Nova Canvas has already stopped responding in the Tokyo region on that basis.

AI Agent Foundry will not add a replacement image generation model, so no image generation model will be supported here once Nova Canvas reaches end of life. For image generation going forward, use the GPT Image model available in [Treasure AI Studio](/products/ai-studio/visualization/image-generation).

**Web search models**

* Open AI GPT-5 mini


Web Search Tool Setup
The Web Search Tool is not visible in the console UI. Configuration is only possible via TDX CLI or Treasure Studio. For detailed setup instructions, see the [Web Search Tool Setup Guide](/treasure-code/guide/web-search-tool) in Treasure Code documentation.

NOTE: Use of any OpenAI model must be limited to the following use cases, and is subject in any case to applicable terms of service and Treasure Data [AI Acceptable Use Policy](https://www.treasuredata.com/terms/ai-acceptable-use-policy/):

* Image Manipulation & Generation (Leveraging gpt-image-1, etc.)
* AI Agent Knowledge Bases & Information Retrieval (Leveraging TD’s Embeddings, etc.)
* Data Analysis & Insights Generation
* User Interface & Accessibility (Leveraging Chat Capabilities)
* Summarization - Use of the services for summarization from trusted source documents such as internal company documentation, but not as an open-ended summarizer without reference to trusted source documents. Examples include summarization of call center transcripts, technical reports, and product reviews.
* Reasoning over structured and unstructured data - Use of the services to analyze inputs using classification, sentiment analysis of text, or entity extraction. Examples include analyzing product feedback sentiment, analyzing support calls and transcripts, and refining text-based search with embeddings.
* Search - Use of the services to search trusted source documents such as internal company documentation, provided that the application does not generate results ungrounded in trusted source documentation.
* Question-answering - Use of the services to ask questions and receive answers from trusted source documents such as internal company documentation, provided that the application does not generate answers ungrounded in trusted source documentation.


If you are interested in using OpenAI models for any use case different from or broader in scope than those described above, reach out to your Account Representative to request prior approval from Treasure Data.