---
schema_version: "1"
id: 6890c62875a8be19df0beb7d4f61090603bfe6f0509c78655d59b456bea297da
title: Large Language Model Limits
collection: admin
version: current
locale: en-us
portal_url: https://docs.salesforce.com/admin/current/en-us/ai/264-0-0/generative_ai_llm_limits.md
canonical_url: https://help.salesforce.com/s/articleView?id=ai.generative_ai_llm_limits.htm&release=264.0.0&type=5
source_site: help.salesforce.com
content_type: documentation
content_hash: 4f958a1ad2c7db279be5419959e68ac47dbf1795557042048cff040b045c74e5
description: Understand limits for supported large language models (LLMs) from multiple providers for embedded features, such as Prompt Builder. Limits for each model include requests per minute and token limits.
products:
  - Agentforce
  - Cross Cloud
release: "264"
---
# Large Language Model Limits

Understand limits for supported large language models (LLMs) from multiple providers for embedded features, such as Prompt Builder. Limits for each model include requests per minute and token limits.

## Rate Limits

There’s a rate limit of 10,000 LLM generation requests per minute (RPM) per org. However, depending on a model’s usage and the available capacity from model providers, the RPM limit can vary by model. For the RPM limit per org for individual models, see the Salesforce-Managed Model Limits section.

For sandboxes, the default rate limit is 500 generation requests per minute per org.

For free and trial orgs, the default rate limits are:

- 1,000 requests per hour
- 5,000,000 tokens per hour
- 1,000,000 tokens per minute

## Salesforce-Managed Model Limits

This table lists the limits for Salesforce-managed models that are available for embedded features, such as Prompt Builder.

     
| **Model Provider** | **Model Family** | **Maximum Requests Per Minute (RPM)** | **Maximum Tokens Per Minute (TPM)**1 | **Maximum Input Tokens** | **Maximum Output Tokens** |
| --- | --- | --- | --- | --- | --- |
| Bedrock (Amazon) | Nova Lite | 500 | 3 million | 300,000 | 5,000 |
| Bedrock (Amazon) | Nova Pro | 500 | 3 million | 300,000 | 5,000 |
| Bedrock (Amazon) | Nova 2 Lite | 1,000 | 2 million | 1,000,000 | 65,536 |
| Bedrock (Anthropic) | Claude Haiku 4.5 | 1,000 | 3 million | 200,000 | 64,000 |
| Bedrock (Anthropic) | Claude Haiku 5.5 | 1,000 | 2 million | 1,000,000 | 128,000 |
| Bedrock (Anthropic) | Claude Opus 4.5 | 1,000 | 4 million | 200,000 | 8,192 |
| Bedrock (Anthropic) | Claude Opus 4.6 | 1,000 | 4 million | 1,000,000 | 128,000 |
| Bedrock (Anthropic) | Claude Opus 4.7 | 1,000 | 4 million | 1,000,000 | 128,000 |
| Bedrock (Anthropic) | Claude Opus 4.8 | 1,000 | 4 million | 1,000,000 | 128,000 |
| Bedrock (Anthropic) | Claude Opus 5 | 1,000 | 2 million | 1,000,000 | 128,000 |
| Bedrock (Anthropic) | Claude Opus 5.5 | 1,000 | 2 million | 1,000,000 | 128,000 |
| Bedrock (Anthropic) | Claude Sonnet 4.5 | 1,000 | 4 million | 200,000 | 64,000 |
| Bedrock (Anthropic) | Claude Sonnet 4.6 | 1,000 | 4 million | 200,000 | 64,000 |
| Bedrock (Anthropic) | Claude Sonnet 5 | 1,000 | 4 million | 200,000 | 64,000 |
| Bedrock (Anthropic) | Claude Sonnet 5.5 | 1,000 | 2 million | 1,000,000 | 128,000 |
| Bedrock (NVIDIA) | Nemotron 3 Nano 30B (Beta) | 1,000 | 5 million | 256,000 | 8,192 |
| Bedrock (NVIDIA) | Nemotron 3 Super 120B (Beta) | 1,000 | 5 million | 256,000 | 32,000 |
| OpenAI and Azure OpenAI | GPT-4o (GPT 4 Omni) | 2,000 | 10 million | 128,000 | 16,384 |
| OpenAI | GPT-4o Mini | 1,000 | 5 million | 128,000 | 16,384 |
| OpenAI and Azure OpenAI | GPT-4o-mini (GPT 4 Omni Mini) | 1,000 | 5 million | 128,000 | 16,384 |
| OpenAI and Azure OpenAI | GPT-4.1 | 2,000 | 10 million | 128,000 | 32,768 |
| OpenAI and Azure OpenAI | GPT-4.1 Mini | 500 | 2 million | 128,000 | 32,768 |
| OpenAI and Azure OpenAI | GPT-5 | 1,000 | 2 million | 272,000 | 128,000 |
| OpenAI and Azure OpenAI | GPT-5 Mini | 1,000 | 2 million | 272,000 | 128,000 |
| OpenAI and Azure OpenAI | GPT 5.1 | 500 | 2 million | 272,000 | 128,000 |
| OpenAI and Azure OpenAI | GPT 5.2 | 500 | 2 million | 272,000 | 128,000 |
| OpenAI and Azure OpenAI | GPT 5.4 | 250 | 2 million | 1,050,000 | 128,000 |
| OpenAI and Azure OpenAI | GPT 5.4 Mini | 1,000 | 2 million | 400,000 | 128,000 |
| OpenAI and Azure OpenAI | GPT 5.5 | 250 | 2 million | 1,050,000 | 128,000 |
| OpenAI and Azure OpenAI | GPT 5.6 Luna | 250 | 5 million | 1,050,000 | 128,000 |
| OpenAI and Azure OpenAI | GPT 5.6 Terra | 250 | 5 million | 1,050,000 | 128,000 |
| OpenAI and Azure OpenAI | GPT 5.6 Sol | 250 | 5 million | 1,050,000 | 128,000 |
| OpenAI and Azure OpenAI | GPT 6 Sol | 1,000 | 2 million | 1,050,000 | 128,000 |
| OpenAI and Azure OpenAI | GPT 6 Luna | 1,000 | 2 million | 1,050,000 | 128,000 |
| OpenAI and Azure OpenAI | GPT 6 Astra | 1,000 | 2 million | 1,050,000 | 128,000 |
| OpenAI and Azure OpenAI | GPT 6.1 Sol | 1,000 | 2 million | 1,000,000 | 131,072 |
| OpenAI and Azure OpenAI | O3 | 500 | 2 million | 200,000 | 100,000 |
| OpenAI and Azure OpenAI | O4 Mini | 500 | 2 million | 200,000 | 100,000 |
| Vertex AI (Google) | Gemini 2.5 Flash | 250 | 2 million | 1,048,576 | 65,536 |
| Vertex AI (Google) | Gemini 2.5 Flash Lite | 250 | 2 million | 1,048,576 | 65,536 |
| Vertex AI (Google) | Gemini 2.5 Pro | 250 | 1 million | 1,048,576 | 65,536 |
| Vertex AI (Google) | Gemini 3 Flash (Beta) | 100 | 2 million | 1,048,576 | 65,536 |
| Vertex AI (Google) | Gemini 3 Pro (Beta) | 50 | 1 million | 1,048,576 | 65,536 |
| Vertex AI (Google) | Gemini 3.1 Flash Lite | 100 | 2 million | 1,048,576 | 65,536 |
| Vertex AI (Google) | Gemini 3.1 Pro (Beta) | 50 | 1 million | 1,048,576 | 65,536 |
| Vertex AI (Google) | Gemini 3.5 Flash | 1,000 | 2 million | 1,048,576 | 65,536 |
| Vertex AI (Google) | Gemini 3.6 Flash | 1,000 | 2 million | 1,048,576 | 65,536 |
| Vertex AI (Google) | Gemini 3.7 Flash | 1,000 | 2 million | 1,048,576 | 65,536 |
| Vertex AI (Google) | Gemini 3.8 Flash | 1,000 | 2 million | 1,048,576 | 65,536 |

1Maximum Tokens Per Minute (TPM) is measured against the sum of input and output tokens.

## Data Masking Token Limit

When data masking is turned on in the Einstein Trust Layer, all models are currently limited to a context size of 65,536 tokens. To turn off data masking and use the full context window, see [Set up Einstein Trust Layer](https://help.salesforce.com/s/articleView?id=ai.generative_ai_trust_setup.htm&release=264.0.0&type=5 "Configure the Einstein Trust Layer to align with your organizational privacy and security policies and requirements.").

#### See Also

- [Large Language Model Support](https://help.salesforce.com/s/articleView?id=ai.generative_ai_large_language_model_support.htm&release=264.0.0&type=5 "Understand supported large language models (LLMs) from multiple providers for embedded features, such as Prompt Builder. Identify the Salesforce-managed models that are available out of the box. Learn how you can bring your own model (BYOLLM) by using AI Models (formerly Einstein Studio).")
- [Large Language Model Multimodal Support](https://help.salesforce.com/s/articleView?id=ai.generative_ai_llm_multimodal_support.htm&release=264.0.0&type=5 "Salesforce-managed models have different levels of support and limits for including JPG, PNG, or PDF files in a model request.")
- [_Agentforce Developer Guide_: Rate Limits for Models API](https://developer.salesforce.com/docs/ai/agentforce/guide/models-api-rate-limits.html)