OpenAI’s GPT-4o mini Now Available in API with Vision and Fine-tuning Text Capabilities on Azure AI (2024)

We recently launched OpenAI’s fastest model, GPT-4o mini, in the Azure OpenAI Studio Playground, simultaneously with OpenAI. The response from our customers has been phenomenal. Today, we are excited to bring this powerful model to even more developers by releasing the GPT-4o mini API with vision support for Global and East US Regional Standard Deployments.

From Playground to API: Expanding Accessibility

Launching GPT-4o mini in the Azure OpenAI Studio Playground provided our customers with the opportunity to experiment and innovate with the latest AI technology. Now, by extending its availability to the API with global and regional pricing, we are empowering developers to seamlessly integrate GPT-4o mini into their applications, leveraging its incredible speed and versatility for a wide range of tasks.

Unlocking New Possibilities with Vision and Text Capabilities

With the addition of vision input capabilities, GPT-4o mini expands its versatility and opens new horizons for developers and businesses. This enhancement allows users to process and analyze visual data, extracting valuable insights and generating comprehensive text outputs. Whether it's interpreting images or processing documents, GPT-4o mini is designed to handle a wide range of tasks and use cases efficiently.

Flexible Pricing: Regional and Global Options

GPT4o-mini is available for Global Standard deployments in all regions and Standard Regional deployments in East US, with more regions coming soon.

Operating costs can vary significantly across different regions due to factors such as data center expenses and local costs for renewable energy. Additionally, the strict compliance and residency requirements offered by Azure necessitate increased infrastructure investments. To provide our customers with the best possible price while maintaining high standards, we are introducing price tiers for regional Standard and Global Standard for GPT-4o mini. Global Standard provides the lowest price with the highest throughput. It is the best starting point for customers without data processing requirements. Regional Standard pricing will fluctuate based on regional operating costs, ensuring that customers receive a fair and transparent pricing model, meeting the requirement of data residency and compliance. This approach aligns with how services like Azure VMs already offer regional pricing, allowing for flexibility and cost-efficiency tailored to specific regional needs.

Key Features and Benefits

1. Enhanced Vision Input: Leverage the power of GPT-4o mini to process images and videos, enabling applications such as visual recognition, scene understanding, and multimedia content analysis.
2. Comprehensive Text Output: Generate detailed and contextually accurate text outputs from visual inputs, making it easier to create reports, summaries, and detailed analyses.
3. Cost-Effective Solutions: Benefit from the cost efficiencies of GPT-4o mini, which is significantly cheaper than previous models, allowing you to deliver high-quality applications at a lower cost. For example, GPT-4o mini offers the quality of GPT-4 Turbo at a price lower than GPT-3.5 Turbo. We are also happy to make the model available in both global and regional standard deployments.

GPT-4o mini Fine Tuning

We are also excited to announce the fine tuning capabilities for GPT-4o mini in public preview. Fine-tuning allows customers to tailor the model to their specific use cases and scenarios, ensuring the outputs are more accurate and relevant.

This update not only enhances our API but also makes GPT-4o mini fine tuning with text content accessible via Azure OpenAI Studio and Azure AI Studio.

Why Should I use GPT-4o mini Fine-Tuning?

GPT-4o mini is smarter and more capable than GPT-3.5 Turbo, is more affordable and offers a longer training context of 64K (4 times that of GPT-3.5 Turbo) and an inference context of 128K (8 times that of GPT-3.5 Turbo).

How much does GPT-4o mini fine tuning cost?

We recently updated our billing for fine tuning with the Azure OpenAI Service to bill based on the number of tokens in your training file – instead of the total elapsed training time. We still charge an hourly rate for deployment with token-based billing for inferencing using the resulting model.

What Else? Enhanced Features and Responsible AI for GPT-4o mini Fine-Tuning

GPT-4o mini supports continuous fine tuning, function calling and tools. You can use continuous fine tuning with GPT-4o mini based model. You can also include function/tool calls in your training data for GPT-4o mini or use function/tool calls with the output model.

And more importantly, safety checks apply to GPT-4o mini fine tuning. We apply the same model safety check for GPT-4o mini as we have in place for GPT-4 and GPT-4o. Before training begins, your data is evaluated for harmful content such as violence, sexual material, hate speech, and self-harm. If detected above a specified severity, the training job will fail, and you'll be notified of the issues so you can adjust your data. After training, the model is assessed for harmful responses using Azure's safety metrics. If the model generates harmful content at an unacceptable rate, you will be informed and can revise your data or use case before resubmitting the fine-tuning job.

Stay Tuned

Resources

Get all the details about GPT-4o mini on Microsoft Learn. New to Azure? Learn more aboutAzure OpenAI Serviceand check out our release newsfeed for thelatest enhancements.