Large Language Model Support

Agentforce Models

This page focuses on supported LLMs for embedded features, such as Prompt Builder. For model options in Agentforce, see Select Agentforce Model Option.

Salesforce-Managed Models

Quickly get started with generative AI features by choosing a Salesforce-managed model. Features like Prompt Builder and the Models API allow you to customize AI implementations with different models and use them in your apps. Salesforce-managed models are enabled by default to speed up the configuration process.

Note If a request fails to process in the designated region(s), Salesforce-Managed Models eventually re-route to a US region to avoid returning an error. You can disable this behavior for some models by turning on "Enable In-Region Model Requests Only" from Einstein Setup. See Geo-Aware LLM Request Routing on the Einstein Generative AI Platform for more info.

This table lists the Salesforce-managed models that are available for embedded features, such as Prompt Builder. For Agentforce, see the Agentforce Models section.

Note All the Salesforce-managed models might not be available in your org.

• In Setup, a Salesforce admin can disable a model provider. See Manage Model Provider Access.
• In AI Models, a Salesforce admin can hide an LLM configuration from being selected in Prompt Builder. See Manage Large Language Model (LLM) Access by Hiding Configurations.

For more details about these supported models, see Supported Models in the Agentforce Developer Guide.

Usage Types

Each model has an assigned usage type, which factors into billing for model usage.

For more details about Usage Type, see Agentforce and Generative AI Usage and Billing.

Inference Profiles

Some Anthropic models are accessible in particular AWS Regions only as a cross-region inference profile. Cross-region inference requests are kept within the AWS regions that are part of the geography where the request originates. For example, a request made within the United States to Claude Sonnet 4 is kept within the AWS Regions in the United States. A request made in Japan can be serviced by any of the destinations in the apac.anthropic.claude-sonnet-4-20250514-v1:0 inference profile.

For more information, see Amazon Bedrock documentation:

• Models at a glance
• Supported Regions and models for inference profiles

For routing for each Salesforce org region for Claude Sonnet 4, see Geo-Aware Routing for Anthropic.

Model Limits

For information about limits per model, such as requests per minute (RPM) and token limits, see Large Language Model Limits.

Model Multimodal Support

For information about multimodel support (JPG, PNG, or PDF) per model, see Large Language Model Multimodal Support.

Beta Models

Beta models are new models from model providers that Salesforce is beta testing. Beta models typically have lower rate limits and may not be available in all regions. A beta model has (Beta) appended to its name. If beta models aren't turned on, they appear as (Disabled) in AI Models.

Note This feature is a pilot or beta service that is subject to the Beta Services Terms at Agreements - Salesforce.com or a written Unified Pilot Agreement if executed by Customer, and applicable terms in the Product Terms Directory. Use of this pilot or beta service is at the Customer's sole discretion.

Before you can turn on beta generative AI models, you must turn on Data 360. Data 360 is automatically provisioned as soon as a Data Cloud license is added to your Salesforce org. See Turn On Data 360.

To turn on beta generative AI models, go to Einstein Setup. After beta models are enabled, you can see them in AI Models and use them just like any Salesforce-managed model.

We recommend that you enable beta models in sandbox or development orgs only.

Bring Your Own Large Language Model (BYOLLM)

The Einstein platform allows you to customize your AI experience by bringing in your own models to Salesforce. You can bring in your own model by using AI Models and write a prompt template in Prompt Builder, which you can then integrate into your own apps or an agent. Some common reasons companies want to use different models with Einstein include:

• Your company has an LLM fine-tuned to your data.
• You can use your Azure, Bedrock, OpenAI, or Vertex account.

BYOLLM supports many of the Salesforce-managed models and these additional models:

To use models and providers not listed on this page, see the LLM Open Connector.

Note For BYOLLM and LLM Open Connector, Salesforce uses a set of IP addresses to communicate with an LLM that you host yourself. If your security policy requires a strict network Access Control List (ACL), make sure to add to your allowlists the IP addresses in the BYO Models and Open Connector IP Addresses section of Salesforce Core Services - IP Addresses and Domains to Allow.

Develop LLM Solutions with the Models API

Developers can use Models API to code custom solutions. See the Models API Developer Guide.

Sustainability

Sustainability is a core value at Salesforce. Selecting the appropriate model is one of the most effective ways to reduce energy consumption, water usage, and carbon emissions. Compare the environmental impact of these models by using the relative Sustainability Score in the bottom-right section of the Agentic Benchmark page on the AI Research site. For more details on Salesforce's approach to AI sustainability, see Sustainability at Salesforce.

Deprecated Models

Model deprecation is the process by which a model provider gradually phases out a model, usually in favor of a new and improved model. A deprecated model may reroute to a preferred model to ensure continuity of service. To learn more, see Prepare for Model Deprecation and Rerouting.

We recommend that you start migrating your applications as soon as the deprecation is announced. During migration, update and test each part of your application with the replacement model that we recommend.

These models are deprecated or rerouted.

Rerouted Models

These models are rerouted.

Announcing New and Deprecated Models

New model announcements and model deprecation announcements are part of the monthly Einstein Platform release notes.