We are pleased to announce the main updates of Azure AI Foundry, our integrated platform for designing, customizing and managing AI applications at the corporate level.
At Microsoft, we are committed to developing AI innovations to strengthen organizations, transformation of industries and redefine productivity. Today we are excited that we can announce the main updates of Azure AI Foundry, our integrated platform for designing, adapting and managing AI application at corporate classes. These updates include breakthrough new models such as Openai’s GPT-4.5, improved techniques of fine fine-tuning and distillation and launching new business tools for agents. Advances to the thesis are designed to speed up the way from experimenting AI to a tangible business impact.
Introduction GPT-4.5 in preview of Azure Openai service
Based on the success of the previous models, the GPT-4.5 is the latest and most favorable model in general Purpus. This research insight shows the success of scaling before and after training, which is a step forward in learning techniques unattended.
- Natural interaction: GPT-4.5 offers for more natural interaction experience. It has a wider knowledge base and its higher “EQ” can help improvise the tasks of coding, writing and solving problems.
- Accuracy and hallucination: With a lower rate of hallucination (37.1% compared to 61.8%) and higher accuracy (62.5% compared to 38.2%) compared to GPT-4O developers can rely on more prior and raison.
- A stronger human alignment: Improved balance techniques improve the ability of GPT-4.5 to follow the instructions, understand nuances and participate in natural conversations, making it a more efficient tool for project coding and management.
Developers can use GPT-4.5 in many ways to increase productivity and creativity. In communication, users can rely on GPT-4.5 to create clear and efficient e-mails, messages and documentation. It also offers personalized learning and coaching experience and helps users to acquire new skills or deepen knowledge in specific areas. During brainstorming sessions, the GPT-4.5 can generate innovative ideas and solutions, making it a valuable tool for creative thinking.
To plan and implement GPT-4.5 projects, it helps users to organize their tasks and ensures thorough and efficient approaches. It can also process complex automation of tasks, simplify complex processes and workflows. Developers can streamline their coding workflows by gradually gaining instructions and recurring automation tasks, save time and shorten errors. Overall, the GPT-4,5 is a versatile model. Starting today, business customers have access to GPT-4.5 in Azure AI Foundry. GPT-4.5, is also available in Github Copilot Chat with Copilot Enterprise Users.
New Models: Phi-4, AI Stability and Recent Edition
The latest wave of AI models shares a common focus: it provides specialized skills with greater efficiency. These editions continue to move towards the purpose of the AI, which excels in specific damaints and requires fewer computing resources. Here are a few standout launching on Azure AI Foundry.
Microsoft Phi models continue to push boys about what is possible with smaller and more efficient architectures:
- Phi-4-Multimodal It unifies the text, speech and vision of interactions knowledge of context. Retail kiosks can now diagnose the product from Via Camera and Voice inputs, eliminating the need for complex manual descriptions.
- Phi-4-mini Impressive packaging of power to only 3.8 billion parameters with a context window of 128 prints. It assumes larger models for coding and mathematical tasks and increases the inference speed by 30% compared to previous models.
AI stability continues to develop generative display using models that accelerate creative workflows:
- Stable 3.5 Diffusion generates fast marketing assets with high loyalty than previous versions, which maintains consistency of brand across different visual styles.
- The stable image of Ultra achieves photorealism for product images and reduces the cost of shooting by accurately rendering material and color loyalty.
- A stable core of the image Improved version of SDXL (generative model AI on image developed by AI stability), high quality output with exceptional speed and efficiency.
Cherere increases the capacity of obtaining information through its latest evaluation technology:
- Cohers Rernnk V3.5 Deliver more accurate search results via your context window 4.096-windows and support for more than 100 languages, lifting the contents of lifting even without the exact match of keywords.
The GPT-4O family expands with two specialized variants:
- The GPT-4-Audio-Prview processes the sound challenge for sound and generates spoken resorts with recognized emotions and emphasis, ideal for digital assistants and customer service.
- The GPT-4O-Realtime-Prview eliminates the delay of conversation with a decrease in breakthrough latency and creates truly human flows.
Agora, a pioneer allowing real-time involvement, uses GPT-4-Realtime.
The GPT-4-Realtime revolutionized voice interaction for our conversational product AI and seized developers with multilingual human voice, stable streaming and ultra-pacific latency across customer service, telemedic and education.
–Patrick Ferriter, Senior Vice President for Product and Marketing, Agora
Together, these advances indicate the development of AI to a more natural, more sensitive and more efficient interactions across various uses and deployment.
Given that our model library exceeds 1,800 offers, we continue to move the boundaries of experimentation and observability. Our new set of fine -tuning tools is complemented by the rise of learning techniques unattended.
- Strengthening: Now in a private preview, this technique teaches models to reason in a new way by rewarding the right logical paths and penalizing incorrect thinking.
- The provided deployment for fine -tuning: The Azure Openai service now offers provided deployment for finely tuned models, which ensures predictable performance and costs through provided permeability (PTU) for token billing.
- Fine -tuning for Mistral Models: Exclusive for Azure AI Foundry, Mistral Large 2411 and Minister 3b now fine tuning for industry -specific tasks such as health care editors.
Secure automation and scale for business agents
In today’s business environment, security and scalabibility are necessary strategic. We introduced two powerful features designed to help you safely use AI for critical tasks:
- Bring your vnet: Agent AI Agent Agent Agent now allows all AI interactions, data processing, and API calls to stay safely within your own virtual network and eliminate public internet exposure. Soon adoptors, such as Fujitsu, use this ability to improve sales by 67% with their sales agent, which saves countless hours that can be redirected to customer commitment and strategic planning, all in holding data integrity.
To learn more about creating applications with multiple agents with Azure AI Foundry, check out this informative webinar.
Create your AI solution using the Azure AI foundry
We are excited about this new development and look forward to using these powerful tools and features to manage innovation. Go with new models and tools in Azure AI Foundry.