Business Software for Microsoft Foundry Agent Service

We are excited to unveil Codestral, our inaugural code generation model. This open-weight generative AI system is specifically crafted for tasks related to code generation, enabling developers to seamlessly write and engage with code via a unified instruction and completion API endpoint. As it becomes proficient in both programming languages and English, Codestral is poised to facilitate the creation of sophisticated AI applications tailored for software developers. With a training foundation that encompasses a wide array of over 80 programming languages—ranging from widely-used options like Python, Java, C, C++, JavaScript, and Bash to more niche languages such as Swift and Fortran—Codestral ensures a versatile support system for developers tackling various coding challenges and projects. Its extensive language capabilities empower developers to confidently navigate different coding environments, making Codestral an invaluable asset in the programming landscape.

Introducing an open-source AI model that can be fine-tuned, distilled, and deployed across various platforms. Our newest instruction-tuned model comes in three sizes: 8B, 70B, and 405B, giving you options to suit different needs. With our open ecosystem, you can expedite your development process using a diverse array of tailored product offerings designed to meet your specific requirements. You have the flexibility to select between real-time inference and batch inference services according to your project's demands. Additionally, you can download model weights to enhance cost efficiency per token while fine-tuning for your application. Improve performance further by utilizing synthetic data and seamlessly deploy your solutions on-premises or in the cloud. Take advantage of Llama system components and expand the model's capabilities through zero-shot tool usage and retrieval-augmented generation (RAG) to foster agentic behaviors. By utilizing 405B high-quality data, you can refine specialized models tailored to distinct use cases, ensuring optimal functionality for your applications. Ultimately, this empowers developers to create innovative solutions that are both efficient and effective.

Mistral Large stands as the premier language model from Mistral AI, engineered for sophisticated text generation and intricate multilingual reasoning tasks such as text comprehension, transformation, and programming code development. This model encompasses support for languages like English, French, Spanish, German, and Italian, which allows it to grasp grammar intricacies and cultural nuances effectively. With an impressive context window of 32,000 tokens, Mistral Large can retain and reference information from lengthy documents with accuracy. Its abilities in precise instruction adherence and native function-calling enhance the development of applications and the modernization of tech stacks. Available on Mistral's platform, Azure AI Studio, and Azure Machine Learning, it also offers the option for self-deployment, catering to sensitive use cases. Benchmarks reveal that Mistral Large performs exceptionally well, securing its position as the second-best model globally that is accessible via an API, just behind GPT-4, illustrating its competitive edge in the AI landscape. Such capabilities make it an invaluable tool for developers seeking to leverage advanced AI technology.

Achieve exceptional response quality through a vector database specifically designed for advanced retrieval augmented generation (RAG) and contemporary search functionalities. Emphasize substantial growth with a robust, enterprise-ready vector database that inherently includes security, compliance, and ethical AI methodologies. Create superior applications utilizing advanced retrieval techniques that are underpinned by years of research and proven customer success. Effortlessly launch your generative AI application with integrated platforms and data sources, including seamless connections to AI models and frameworks. Facilitate the automatic data upload from an extensive array of compatible Azure and third-party sources. Enhance vector data processing with comprehensive features for extraction, chunking, enrichment, and vectorization, all streamlined in a single workflow. Offer support for diverse vector types, hybrid models, multilingual capabilities, and metadata filtering. Go beyond simple vector searches by incorporating keyword match scoring, reranking, geospatial search capabilities, and autocomplete features. This holistic approach ensures that your applications can meet a wide range of user needs and adapt to evolving demands.

The latest iteration of the open-source AI model, which can be fine-tuned and deployed in various environments, is now offered in multiple versions, including 1B, 3B, 11B, and 90B, alongside the option to continue utilizing Llama 3.1. Llama 3.2 comprises a series of large language models (LLMs) that come pretrained and fine-tuned in 1B and 3B configurations for multilingual text only, while the 11B and 90B models accommodate both text and image inputs, producing text outputs. With this new release, you can create highly effective and efficient applications tailored to your needs. For on-device applications, such as summarizing phone discussions or accessing calendar tools, the 1B or 3B models are ideal choices. Meanwhile, the 11B or 90B models excel in image-related tasks, enabling you to transform existing images or extract additional information from images of your environment. Overall, this diverse range of models allows developers to explore innovative use cases across various domains.

The newest version in the Llama series, Llama 3.3, represents a significant advancement in language models aimed at enhancing AI's capabilities in understanding and communication. It boasts improved contextual reasoning, superior language generation, and advanced fine-tuning features aimed at producing exceptionally accurate, human-like responses across a variety of uses. This iteration incorporates a more extensive training dataset, refined algorithms for deeper comprehension, and mitigated biases compared to earlier versions. Llama 3.3 stands out in applications including natural language understanding, creative writing, technical explanations, and multilingual interactions, making it a crucial asset for businesses, developers, and researchers alike. Additionally, its modular architecture facilitates customizable deployment in specific fields, ensuring it remains versatile and high-performing even in large-scale applications. With these enhancements, Llama 3.3 is poised to redefine the standards of AI language models.

MacWhisper allows users to efficiently convert audio content into written text by harnessing OpenAI's Whisper technology. Users have the option to record audio directly from their microphone or any compatible input device on their Mac, or they can simply drag and drop audio files for precise transcription. It is capable of capturing meetings from various platforms, including Zoom, Teams, Webex, Skype, Chime, and Discord, while ensuring that all transcription is processed locally to maintain user privacy. Transcripts generated can be saved or exported in several formats, such as .srt, .vtt, .csv, .docx, .pdf, markdown, and HTML. MacWhisper is known for its rapid transcription capabilities, supporting over 100 languages, and features like transcript searching, synchronized audio playback, removal of filler words, and the ability to add speaker labels. The Pro version further extends its offerings with features like batch transcription, the ability to transcribe YouTube videos, integrations with AI services such as OpenAI's ChatGPT and Anthropic's Claude, as well as system-wide dictation and translation options for audio files into different languages. This makes MacWhisper an exceptional tool not just for individuals but also for professionals who require versatile transcription solutions.

Designed on a container-based runtime, this platform enhances scalability and portability while automating essential business processes across any location. Upgrade your BizTalk Server applications by transitioning them to Logic Apps with the help of the BizTalk migration tool. Establish connections between logic apps and your virtual networks to integrate cloud-based and on-premises solutions in a smooth and secure manner. By containerizing your workflows, you can deploy and operate your applications in any environment—be it cloud, on-premises, or your preferred infrastructure. Implement CI/CD best practices to optimize your workflows and utilize built-in tools to ensure seamless and secure deployments. Logic applications can be deployed and executed in Azure, across various containers, and on local systems. Additionally, you can enable private endpoints, simplify virtual network access, and utilize deployment slots for better management. Develop, debug, and test applications on Windows, MacOS, and Linux using Visual Studio Code for maximum flexibility. By deploying multiple workflows within a single logic app, you can streamline automated deployments and enhance your CI/CD pipelines, making the overall process more efficient and manageable. Overall, this approach not only boosts productivity but also ensures that your applications are consistently up to date.

Introducing the next iteration of our open-source large language model, this version features model weights along with initial code for the pretrained and fine-tuned Llama language models, which span from 7 billion to 70 billion parameters. The Llama 2 pretrained models have been developed using an impressive 2 trillion tokens and offer double the context length compared to their predecessor, Llama 1. Furthermore, the fine-tuned models have been enhanced through the analysis of over 1 million human annotations. Llama 2 demonstrates superior performance against various other open-source language models across multiple external benchmarks, excelling in areas such as reasoning, coding capabilities, proficiency, and knowledge assessments. For its training, Llama 2 utilized publicly accessible online data sources, while the fine-tuned variant, Llama-2-chat, incorporates publicly available instruction datasets along with the aforementioned extensive human annotations. Our initiative enjoys strong support from a diverse array of global stakeholders who are enthusiastic about our open approach to AI, including companies that have provided valuable early feedback and are eager to collaborate using Llama 2. The excitement surrounding Llama 2 signifies a pivotal shift in how AI can be developed and utilized collectively.

Pixtral Large is an expansive multimodal model featuring 124 billion parameters, crafted by Mistral AI and enhancing their previous Mistral Large 2 framework. This model combines a 123-billion-parameter multimodal decoder with a 1-billion-parameter vision encoder, allowing it to excel in the interpretation of various content types, including documents, charts, and natural images, all while retaining superior text comprehension abilities. With the capability to manage a context window of 128,000 tokens, Pixtral Large can efficiently analyze at least 30 high-resolution images at once. It has achieved remarkable results on benchmarks like MathVista, DocVQA, and VQAv2, outpacing competitors such as GPT-4o and Gemini-1.5 Pro. Available for research and educational purposes under the Mistral Research License, it also has a Mistral Commercial License for business applications. This versatility makes Pixtral Large a valuable tool for both academic research and commercial innovations.

Kong AI Gateway serves as a sophisticated semantic AI gateway that manages and secures traffic from Large Language Models (LLMs), facilitating the rapid integration of Generative AI (GenAI) through innovative semantic AI plugins. This platform empowers users to seamlessly integrate, secure, and monitor widely-used LLMs while enhancing AI interactions with features like semantic caching and robust security protocols. Additionally, it introduces advanced prompt engineering techniques to ensure compliance and governance are maintained. Developers benefit from the simplicity of adapting their existing AI applications with just a single line of code, which significantly streamlines the migration process. Furthermore, Kong AI Gateway provides no-code AI integrations, enabling users to transform and enrich API responses effortlessly through declarative configurations. By establishing advanced prompt security measures, it determines acceptable behaviors and facilitates the creation of optimized prompts using AI templates that are compatible with OpenAI's interface. This powerful combination of features positions Kong AI Gateway as an essential tool for organizations looking to harness the full potential of AI technology.

Orq.ai stands out as the leading platform tailored for software teams to effectively manage agentic AI systems on a large scale. It allows you to refine prompts, implement various use cases, and track performance meticulously, ensuring no blind spots and eliminating the need for vibe checks. Users can test different prompts and LLM settings prior to launching them into production. Furthermore, it provides the capability to assess agentic AI systems within offline environments. The platform enables the deployment of GenAI features to designated user groups, all while maintaining robust guardrails, prioritizing data privacy, and utilizing advanced RAG pipelines. It also offers the ability to visualize all agent-triggered events, facilitating rapid debugging. Users gain detailed oversight of costs, latency, and overall performance. Additionally, you can connect with your preferred AI models or even integrate your own. Orq.ai accelerates workflow efficiency with readily available components specifically designed for agentic AI systems. It centralizes the management of essential phases in the LLM application lifecycle within a single platform. With options for self-hosted or hybrid deployment, it ensures compliance with SOC 2 and GDPR standards, thereby providing enterprise-level security. This comprehensive approach not only streamlines operations but also empowers teams to innovate and adapt swiftly in a dynamic technological landscape.

Enhance your development process with Functions, a serverless compute platform designed for event-driven applications that can tackle intricate orchestration challenges. You can efficiently build and troubleshoot your applications locally without requiring extra configuration, and easily deploy and manage them at scale in the cloud while utilizing triggers and bindings for service integration. Enjoy a comprehensive development experience that comes with integrated tools and built-in DevOps features. The platform offers a unified programming model that enables you to respond to events and effortlessly connect with various services. You can create a range of functions and use cases, including web applications and APIs using .NET, Node.js, or Java; machine learning processes through Python; and automate cloud tasks with PowerShell. This approach provides a holistic serverless application development journey—from local construction and debugging to cloud deployment and monitoring, ensuring a seamless transition at every stage. With such capabilities, developers can maximize their productivity and innovation potential.

Orate is a comprehensive AI toolkit designed for speech that empowers developers to generate lifelike, human-like audio and transcribe spoken language through a cohesive API that works with major AI platforms including OpenAI, ElevenLabs, and AssemblyAI. This platform features text-to-speech capabilities, allowing users to effortlessly convert written text into realistic audio by utilizing a user-friendly API that integrates with multiple service providers. For example, developers can easily generate speech from text prompts by importing the 'speak' function from Orate alongside their selected provider. Furthermore, Orate excels in speech-to-text processing, converting spoken words into accurate and meaningful text with exceptional speed and dependability. By utilizing the 'transcribe' function in conjunction with the desired provider, users can efficiently convert audio files into written content. Additionally, the toolkit includes features for speech-to-speech conversions, allowing users to modify the voice in their audio with a straightforward voice-to-voice API that is compatible with leading AI services, thereby offering a versatile solution for various audio processing needs. With its broad range of functionalities, Orate stands out as a powerful tool for anyone looking to enhance their audio applications.

Aurascape is a cutting-edge security platform tailored for the AI era, empowering businesses to innovate securely amidst the rapid advancements of artificial intelligence. It offers an all-encompassing view of interactions between AI applications, effectively protecting against potential data breaches and threats driven by AI technologies. Among its standout features are the ability to oversee AI activity across a wide range of applications, safeguarding sensitive information to meet compliance standards, defending against zero-day vulnerabilities, enabling the secure implementation of AI copilots, establishing guardrails for coding assistants, and streamlining AI security workflows through automation. The core mission of Aurascape is to foster a confident adoption of AI tools within organizations while ensuring strong security protocols are in place. As AI applications evolve, their interactions become increasingly dynamic, real-time, and autonomous, necessitating robust protective measures. By preempting emerging threats, safeguarding data with exceptional accuracy, and enhancing team productivity, Aurascape also monitors unauthorized app usage, identifies risky authentication practices, and curtails unsafe data sharing. This comprehensive security approach not only mitigates risks but also empowers organizations to fully leverage the potential of AI technologies.

Phi-4-mini-flash-reasoning is a 3.8 billion-parameter model that is part of Microsoft's Phi series, specifically designed for edge, mobile, and other environments with constrained resources where processing power, memory, and speed are limited. This innovative model features the SambaY hybrid decoder architecture, integrating Gated Memory Units (GMUs) with Mamba state-space and sliding-window attention layers, achieving up to ten times the throughput and a latency reduction of 2 to 3 times compared to its earlier versions without compromising on its ability to perform complex mathematical and logical reasoning. With a support for a context length of 64K tokens and being fine-tuned on high-quality synthetic datasets, it is particularly adept at handling long-context retrieval, reasoning tasks, and real-time inference, all manageable on a single GPU. Available through platforms such as Azure AI Foundry, NVIDIA API Catalog, and Hugging Face, Phi-4-mini-flash-reasoning empowers developers to create applications that are not only fast but also scalable and capable of intensive logical processing. This accessibility allows a broader range of developers to leverage its capabilities for innovative solutions.

Llama (Large Language Model Meta AI) stands as a cutting-edge foundational large language model aimed at helping researchers push the boundaries of their work within this area of artificial intelligence. By providing smaller yet highly effective models like Llama, the research community can benefit even if they lack extensive infrastructure, thus promoting greater accessibility in this dynamic and rapidly evolving domain. Creating smaller foundational models such as Llama is advantageous in the landscape of large language models, as it demands significantly reduced computational power and resources, facilitating the testing of innovative methods, confirming existing research, and investigating new applications. These foundational models leverage extensive unlabeled datasets, making them exceptionally suitable for fine-tuning across a range of tasks. We are offering Llama in multiple sizes (7B, 13B, 33B, and 65B parameters), accompanied by a detailed Llama model card that outlines our development process while adhering to our commitment to Responsible AI principles. By making these resources available, we aim to empower a broader segment of the research community to engage with and contribute to advancements in AI.