Salesforce-managed models have different levels of support and limits for including JPG, PNG, or PDF files in a model request.
Request Limits
To ensure quality of service, we have some limits for multimodal model requests. These limits apply to all Salesforce-managed models.
- Maximum total file size per request: 15MB
- Maximum number of files per request: 10 (any combination of JPG, PNG, or PDF)
- Maximum number of PDF files per request for Bedrock: 5
- Multimodal requests have a 300 second timeout window. Requests longer than 300 seconds fail.
Model Support
This table lists multimodal support for Salesforce-managed models.
| Model Provider | Model Family | Image (JPG or PNG) | |
|---|---|---|---|
| Bedrock (Amazon) | Nova Lite | Yes | Yes |
| Bedrock (Amazon) | Nova Pro | Yes | Yes |
| Bedrock (Amazon) | Nova 2 Lite | Yes | Yes |
| Bedrock (Anthropic) | Claude Haiku 4.5 | Yes | Yes |
| Bedrock (Anthropic) | Claude Haiku 5.5 | Yes | Yes |
| Bedrock (Anthropic) | Claude Opus 4.5 | Yes | Yes |
| Bedrock (Anthropic) | Claude Opus 4.6 | Yes | Yes |
| Bedrock (Anthropic) | Claude Opus 4.7 | Yes | Yes |
| Bedrock (Anthropic) | Claude Opus 4.8 | Yes | Yes |
| Bedrock (Anthropic) | Claude Opus 5 | Yes | Yes |
| Bedrock (Anthropic) | Claude Opus 5.5 | Yes | Yes |
| Bedrock (Anthropic) | Claude Sonnet 4.5 | Yes | Yes |
| Bedrock (Anthropic) | Claude Sonnet 4.6 | Yes | Yes |
| Bedrock (Anthropic) | Claude Sonnet 5 | Yes | Yes |
| Bedrock (Anthropic) | Claude Sonnet 5.5 | Yes | Yes |
| Bedrock (NVIDIA) | Nemotron 3 Nano 30B (Beta) | No | No |
| Bedrock (NVIDIA) | Nemotron 3 Super 120B (Beta) | No | No |
| OpenAI and Azure OpenAI | GPT-4o (GPT 4 Omni) | Yes | No1 |
| OpenAI | GPT-4o Mini | Yes | Yes |
| OpenAI and Azure OpenAI | GPT-4o-mini (GPT 4 Omni Mini) | Yes | No |
| OpenAI | GPT 4.1 | Yes | Yes |
| Azure OpenAI | GPT 4.1 | Yes | No |
| OpenAI and Azure OpenAI | GPT 4.1 Mini | Yes | No |
| OpenAI and Azure OpenAI | GPT-5 | Yes | Yes |
| OpenAI and Azure OpenAI | GPT 5.1 | Yes | Yes |
| OpenAI and Azure OpenAI | GPT 5.2 | Yes | Yes |
| OpenAI and Azure OpenAI | GPT 5.4 | Yes | Yes |
| OpenAI and Azure OpenAI | GPT 5.4 Mini | Yes | Yes |
| OpenAI and Azure OpenAI | GPT 5.5 | Yes | Yes |
| OpenAI and Azure OpenAI | GPT-5 Mini | Yes | Yes |
| OpenAI and Azure OpenAI | GPT 5.6 Luna | Yes | No |
| OpenAI and Azure OpenAI | GPT 5.6 Terra | Yes | No |
| OpenAI and Azure OpenAI | GPT 5.6 Sol | Yes | No |
| OpenAI and Azure OpenAI | GPT 6 Sol | Yes | Yes |
| OpenAI and Azure OpenAI | GPT 6 Luna | Yes | Yes |
| OpenAI and Azure OpenAI | GPT 6 Astra | Yes | Yes |
| OpenAI and Azure OpenAI | GPT 6.1 Sol | Yes | No |
| OpenAI and Azure OpenAI | O3 | Yes | No |
| OpenAI and Azure OpenAI | O4 | Yes | No |
| Vertex AI (Google) | Gemini 2.5 Flash | Yes | Yes |
| Vertex AI (Google) | Gemini 2.5 Flash Lite | Yes | Yes |
| Vertex AI (Google) | Gemini 2.5 Pro | Yes | Yes |
| Vertex AI (Google) | Gemini 3 Flash (Beta) | Yes | Yes |
| Vertex AI (Google) | Gemini 3 Pro (Beta) | Yes | Yes |
| Vertex AI (Google) | Gemini 3.1 Flash Lite | Yes | Yes |
| Vertex AI (Google) | Gemini 3.1 Pro (Beta) | Yes | Yes |
| Vertex AI (Google) | Gemini 3.5 Flash | Yes | Yes |
| Vertex AI (Google) | Gemini 3.6 Flash | Yes | Yes |
| Vertex AI (Google) | Gemini 3.7 Flash | Yes | Yes |
| Vertex AI (Google) | Gemini 3.8 Flash | Yes | Yes |
1 A geo-aware model, such as GPT 4 Omni, uses OpenAI or Azure OpenAI to service LLM requests. Because Azure OpenAI doesn't support PDF input, you can't send a PDF in a request to a geo-aware model. For more information, see Geo-Aware LLM Request Routing on the Einstein Generative AI Platform.
Model Provider File Size Limits
Some model providers have extra limits for individual file size. The next table gives more details on these extra file size limits.
| Model Provider | File Size Limits |
|---|---|
| Bedrock (Amazon and Anthropic) |
|
| Vertex AI (Google) | 7 MB per JPG or PNG |
Einstein Trust Layer Support
If a PDF is uploaded as a binary file in a model request, Einstein Trust Layer protections are not applied. To apply Einstein Trust Layer protections, you must extract the contents of the PDF and send the contents in a prompt in text format. Similarly, Einstein Trust Layer protections are not applied to JPG or PNG images.