Vision-enabled chat model concepts

Vision-enabled chat models are large multimodal models (LMM) developed by OpenAI that can analyze images and provide textual responses to questions about them. They incorporate both natural language processing and visual understanding. The current vision-enabled models are GPT-4 Turbo with Vision, GPT-4o, and GPT-4o-mini. This guide provides details on their capabilities and limitations.

To try out vision-enabled chat models, see the quickstart.

Vision-enabled chats

The vision-enabled models answer general questions about what's present in the images you upload.

Special pricing information

Svarīgi

Pricing details are subject to change in the future.

Vision-enabled models accrue charges like other Azure OpenAI chat models. You pay a per-token rate for the prompts and completions, detailed on the Pricing page. The base charges and additional features are outlined here:

Base Pricing for GPT-4 Turbo with Vision is:

• Input: $0.01 per 1000 tokens
• Output: $0.03 per 1000 tokens

See the Tokens section of the overview for information on how text and images translate to tokens.

Example image price calculation

Svarīgi

The following content is an example only, and prices are subject to change in the future.

For a typical use case, take an image with both visible objects and text and a 100-token prompt input. When the service processes the prompt, it generates 100 tokens of output. In the image, both text and objects can be detected. The price of this transaction would be:

Item	Detail	Cost
Text prompt input	100 text tokens	$0.001
Example image input (see Image tokens)	170 + 85 image tokens	$0.00255
Enhanced add-on features for OCR	$1.50 / 1000 transactions	$0.0015
Enhanced add-on features for Object Grounding	$1.50 / 1000 transactions	$0.0015
Output Tokens	100 tokens (assumed)	$0.003
Total		$0.00955

Input limitations

This section describes the limitations of vision-enabled chat models.

Image support

• Maximum input image size: The maximum size for input images is restricted to 20 MB.
• Low resolution accuracy: When images are analyzed using the "low resolution" setting, it allows for faster responses and uses fewer input tokens for certain use cases. However, this could impact the accuracy of object and text recognition within the image.
• Image chat restriction: When you upload images in Azure AI Foundry portal or the API, there is a limit of 10 images per chat call.

Next steps

• Get started using vision-enabled models by following the quickstart.
• For a more in-depth look at the APIs, follow the how-to guide.
• See the completions and embeddings API reference