On April 16, 2025, Microsoft released new AI models on Azure AI Foundry with significantly improved capabilities for coding, instruction following, and long-context processing. This article looks at these new models in detail to help you with the decision on what models to use within your organization.
Background
Microsoft has introduced new OpenAI models, including GPT-4.1, GPT-4.1-mini, GPT-4.1-nano, o3, and o4-mini. These releases are part of a broader effort to provide more advanced and versatile AI tools in Azure. The new models are especially strong for coding applications and allow them to process longer contexts, which makes them highly useful for developers and businesses alike. For example, GPT-4.1 delivers functional code with fewer errors and is particularly effective in single-shot tasks requiring precision and efficiency.
Comparing GPT-4.1, o3, o4-mini and GPT-4o variations
Here is a comparison table for the new AI models available in Azure AI Foundry:
| Feature/Model | GPT-4.1 | GPT-4.1-mini | GPT-4.1-nano | o3 | o4-mini |
| Description | Latest iteration of GPT-4o, optimized for coding and instruction tasks | Balanced version of GPT-4.1 | Cost-efficient version of GPT-4.1 | Advanced AI reasoning model | Compact version of OpenAI o3 |
| Strengths | Enhanced coding and instruction following, long context support | Balanced performance and cost, long context support | Lowest cost, long context support | Enhanced quality, safety, and performance, multimodality support | Improved safety and performance, vision analysis capabilities |
| Weaknesses | Higher cost | Less accurate than GPT-4.1 | Lower reasoning and accuracy | Higher cost | Less powerful than OpenAI o3 |
| Usage Examples | Complex Coding Tasks – Developing a complex algorithm for a machine learning model. | Quick Prototyping – Creating a prototype for a new web application. | Low-Latency Applications – Real-time text autocompletion. | Detailed Analytical Tasks – Conducting a comprehensive market analysis. | Fast and Efficient Reasoning – Solving complex mathematical problems. |
| Reasoning & Accuracy | Highest | Balanced | Lower | High | High |
| Cost & Efficiency | Higher cost | Balanced | Lowest cost | Higher cost | Balanced |
| Quality index | 0.9 | 0.89 | Not specified | 0.81 | 0.7 |
| Other Categories | Supports tool calling and structured outputs | Supports tool calling and structured outputs | Supports tool calling and structured outputs | Supports multiple APIs, reasoning summary, and parallel tool calling | Supports multiple APIs, reasoning summary, and parallel tool calling |
More Information
Find info on how to license Azure and save costs in our article here: https://www.schneider.im/microsoft-azure-licensing-full-guide/.
Announcements: https://azure.microsoft.com/en-us/blog/announcing-the-gpt-4-1-model-series-for-azure-ai-foundry-developers/ and https://azure.microsoft.com/en-us/blog/o3-and-o4-mini-unlock-enterprise-agent-workflows-with-next-level-reasoning-ai-with-azure-ai-foundry-and-github/.
Ready to get started in Azure?
Get expert Azure licensing consultancy and save on Azure with SCHNEIDER IT MANAGEMENT. Contact us to get started.
