News from the AI & ML world

DeeperML - #phi-4

@www.microsoft.com //

Microsoft Advances AI Research and Fusion Energy - Microsoft is advancing AI technologies, releasing Phi-4-reasoning, and introducing ARTIST (Agentic Reasoning and Tool Integration in Self-improving Transformers), a framework that combines agentic reasoning, reinforcement learning, and dynamic tool use to enhance LLMs.
Microsoft is pushing the boundaries of AI with advancements in both model efficiency and novel applications. The company recently commemorated the one-year anniversary of Phi-3 by introducing three new small language models: Phi-4-reasoning, Phi-4-reasoning-plus, and Phi-4-mini-reasoning. These models are designed to deliver complex reasoning capabilities that rival much larger models while maintaining efficiency for diverse computing environments. According to Microsoft, "Phi-4-reasoning generates detailed reasoning chains that effectively leverage additional inference-time compute," demonstrating that high-quality synthetic data and careful curation can lead to smaller models that perform comparably to their more powerful counterparts.

The 14-billion parameter Phi-4-reasoning and its enhanced version, Phi-4-reasoning-plus, have shown outstanding performance on numerous benchmarks, outperforming larger models. Notably, they achieve better results than OpenAI's o1-mini and a DeepSeek R1 distill on Llama 70B on mathematical reasoning and PhD-level science questions. Furthermore, Phi-4-reasoning-plus surpasses the massive 671-billion parameter DeepSeek-R1 model on AIME and HMMT evaluations. These results highlight the efficiency and competitive edge of the new models.

In addition to pushing efficiency, Microsoft Research has introduced ARTIST (Agentic Reasoning and Tool Integration in Self-improving Transformers), a framework that combines agentic reasoning, reinforcement learning, and dynamic tool use to enhance LLMs. ARTIST enables models to autonomously decide when, how, and which tools to use. This framework aims to address the limitations of static internal knowledge and text-only reasoning, especially in tasks requiring real-time information or domain-specific expertise. The integration of reinforcement learning allows the models to adapt dynamically and interact with external tools and environments during the reasoning process, ultimately improving their performance in real-world applications.

Recommended read:
References :
  • Microsoft Research: In this issue: New research on compound AI systems and causal verification of the Confidential Consortium Framework; release of Phi-4-reasoning; enriching tabular data with semantic structure, and more. The post appeared first on .
  • www.microsoft.com: Research Focus: Week of May 7, 2025
  • learn.aisingapore.org: Phi-4-reasoning, a 14-billion parameter model, has been released by Microsoft. The model has shown promise in achieving competitive performance with larger models through supervised fine-tuning and synthetic data curation.
  • Source: Microsoft Fusion Summit explores how AI can accelerate fusion research
Ellie Ramirez-Camara@Data Phoenix //

Microsoft's Phi-4 Models Enhance AI Reasoning Performance - Microsoft enhances its Phi-4 family with reasoning models and integrates the Agent2Agent (A2A) protocol to Azure AI Foundry and Copilot Studio for cross-platform AI agent collaboration, while developer Simon Willison notes the Phi-4 model's extensive reasoning output.
Microsoft is expanding its AI capabilities with enhancements to its Phi-4 family and the integration of the Agent2Agent (A2A) protocol. The company's new Phi-4-Reasoning and Phi-4-Reasoning-Plus models are designed to deliver strong reasoning performance with low latency. In addition, Microsoft is embracing interoperability by adding support for the open A2A protocol to Azure AI Foundry and Copilot Studio. This move aims to facilitate seamless collaboration between AI agents across various platforms, fostering a more connected and efficient AI ecosystem.

Microsoft's integration of the A2A protocol into Azure AI Foundry and Copilot Studio will empower AI agents to work together across platforms. The A2A protocol defines how agents formulate tasks and execute them, enabling them to delegate tasks, share data, and act together. With A2A support, Copilot Studio agents can call on external agents, including those outside the Microsoft ecosystem and built with tools like LangChain or Semantic Kernel. Microsoft reports that over 230,000 organizations are already utilizing Copilot Studio, with 90 percent of the Fortune 500 among them. Developers can now access sample applications demonstrating automated meeting scheduling between agents.

Independant developer Simon Willison has been testing the phi4-reasoning model, and reported that the 11GB download (available via Ollama) may well overthink things. Willison noted that it produced 56 sentences of reasoning output in response to a prompt of "hi". Microsoft is actively contributing to the A2A specification work on GitHub and intends to play a role in driving its future development. A public preview of A2A in Azure Foundry and Copilot Studio is anticipated to launch soon. Microsoft envisions protocols like A2A as the bedrock of a novel software architecture where interconnected agents automate daily workflows and collaborate across platforms with auditability and control.

Recommended read:
References :
  • bsky.app: Microsoft's phi4-reasoning model, an 11GB download (via Ollama) which may well overthink things
  • Simon Willison: Simon Willison Published some notes on Microsoft's phi4-reasoning model
  • the-decoder.com: Microsoft leverages Google's open A2A protocol for interoperable AI agents
  • the-decoder.com: Microsoft's Phi 4 responds to a simple "Hi" with 56 thoughts
  • Data Phoenix: Microsoft has introduced three new small language models—Phi-4-reasoning, Phi-4-reasoning-plus, and Phi-4-mini-reasoning—that reportedly deliver complex reasoning capabilities comparable to much larger models while maintaining efficiency for deployment across various computing environments.
  • www.microsoft.com: In this issue: New research on compound AI systems and causal verification of the Confidential Consortium Framework; release of Phi-4-reasoning; enriching tabular data with semantic structure, and more.
Carl Franzen@AI News | VentureBeat //

Microsoft Unveils Phi-4 Reasoning SLM Models Advancing AI - Microsoft unveiled the Phi-4 reasoning SLM models, optimized for complex reasoning tasks like math and coding, outperforming larger models on limited resources.
Microsoft has recently launched its Phi-4 reasoning models, marking a significant stride in the realm of small language models (SLMs). This expansion of the Phi series includes three new variants: Phi-4-reasoning, Phi-4-reasoning-plus, and Phi-4-mini-reasoning, designed to excel in advanced reasoning tasks like mathematics and coding. The company's new models are optimized for complex problems, and can handle complex problems through structured reasoning and internal reflection, while remaining lightweight enough to run on lower-end hardware, including mobile devices.

Microsoft asserts that these models demonstrate that smaller AI can achieve impressive results, rivaling much larger models while operating efficiently on devices with limited resources. CEO Satya Nadella says Microsoft's AI model performance is "doubling every 6 months" due to pre-training, inference, and system design. The Phi-4-reasoning model contains 14 billion parameters and was trained via supervised fine-tuning using reasoning paths from OpenAI's o3-mini. A more advanced version, Phi-4-reasoning-plus, adds reinforcement learning and processes 1.5 times more tokens than the base model.

These new models leverage distillation, reinforcement learning, and high-quality data to achieve their performance. In a demonstration, the Phi-4-reasoning model correctly solved a wordplay riddle by recognizing patterns and applying local reasoning, showcasing its ability to identify patterns, understand riddles, and perform mathematical operations. Despite having just 14 billion parameters, the Phi-4 reasoning models match or outperform significantly larger systems, including the 70B parameter DeepSeek-R1-Distill-Llama. On the AIME-2025 benchmark, the Phi models also surpass DeepSeek-R1, which has 671 billion parameters.

Recommended read:
References :
  • Ken Yeung: Microsoft is doubling down on small language models with new Phi-4 variants that aim to prove a bold idea: small AI can think big.
  • www.windowscentral.com: Microsoft just launched expanded small language models (SLMs) based on its own Phi-4 AI.
  • THE DECODER: Microsoft is expanding its Phi series of compact language models with three new variants designed for advanced reasoning tasks.
  • the-decoder.com: Microsoft's Phi 4 responds to a simple "Hi" with 56 thoughts
  • Data Phoenix: Microsoft launches Phi-4 'reasoning' models to celebrate Phi-3's first anniversary
Matthias Bastian@THE DECODER //

Microsoft's Phi-4 Reasoning Models - Microsoft launched new Phi-4 reasoning models designed to excel at complex reasoning tasks, demonstrating that smaller AI models can achieve significant performance.
References: THE DECODER , Ken Yeung , Analytics Vidhya ...
Microsoft has launched three new additions to its Phi series of compact language models: Phi-4-reasoning, Phi-4-reasoning-plus, and Phi-4-mini-reasoning. These models are designed to excel in complex reasoning tasks, including mathematical problem-solving, algorithmic planning, and coding, demonstrating that smaller AI models can achieve significant performance. The models are optimized to handle complex problems through structured reasoning and internal reflection, while also being efficient enough to run on lower-end hardware, including mobile devices, making advanced AI accessible on resource-limited devices.

Phi-4-reasoning, a 14-billion parameter model, was trained using supervised fine-tuning with reasoning paths from OpenAI's o3-mini. Phi-4-reasoning-plus enhances this with reinforcement learning and processes more tokens, leading to higher accuracy, although with increased computational cost. Notably, these models outperform larger systems, such as the 70B parameter DeepSeek-R1-Distill-Llama, and even surpass DeepSeek-R1 with 671 billion parameters on the AIME-2025 benchmark, a qualifier for the U.S. Mathematical Olympiad, highlighting the effectiveness of Microsoft's approach to efficient, high-performing AI.

The Phi-4 reasoning models show strong results in programming, algorithmic problem-solving, and planning tasks, with improvements in logical reasoning positively impacting general capabilities such as following prompts and answering questions based on long-form content. Microsoft employed a data-centric training strategy, using structured reasoning outputs marked with special tokens to guide the model's intermediate reasoning steps. The open-weight models have been released with transparent training details and are hosted on Hugging Face, allowing for public access, fine-tuning, and use in various applications under a permissive MIT license.

Recommended read:
References :
  • THE DECODER: Microsoft is expanding its Phi series of compact language models with three new variants designed for advanced reasoning tasks.
  • Ken Yeung: Microsoft’s New Phi-4 Variants Show Just How Far Small AI Can Go
  • AI News | VentureBeat: Microsoft Research has announced the release of Phi-4-reasoning-plus, an open-weight language model built for tasks requiring deep, structured reasoning.
  • Analytics Vidhya: Microsoft isn’t like OpenAI, Google, and Meta; especially not when it comes to large language models.
  • MarkTechPost: Despite notable advancements in large language models (LLMs), effective performance on reasoning-intensive tasks—such as mathematical problem solving, algorithmic planning, or coding—remains constrained by model size, training methodology, and inference-time capabilities.
  • the-decoder.com: Microsoft's Phi-4-reasoning models outperform larger models and run on your laptop or phone
  • www.tomsguide.com: Microsoft just unveiled new Phi-4 reasoning AI models — here's why they're a big deal
  • www.marktechpost.com: Despite notable advancements in large language models (LLMs), effective performance on reasoning-intensive tasks—such as mathematical problem solving, algorithmic planning, or coding—remains constrained by model size, training methodology, and inference-time capabilities.
  • www.windowscentral.com: Microsoft just launched expanded small language models (SLMs) based on its own Phi-4 AI.
  • simonwillison.net: This article discusses Microsoft's phi4-reasoning model, which generates 56 sentences of reasoning output in response to a simple prompt.
  • Data Phoenix: Microsoft launches Phi-4 'reasoning' models to celebrate Phi-3's first anniversary
Carl Franzen@AI News | VentureBeat //

Microsoft Launches Phi-4 Reasoning Plus: Small, Powerful Model - Microsoft has launched Phi-4-Reasoning-Plus, a small language model designed for advanced reasoning tasks, with Microsoft CEO revealing that AI is now contributing to 30% of Microsoft's code.
Microsoft has announced the release of Phi-4-reasoning-plus, a new small, open-weight language model designed for advanced reasoning tasks. Building upon the architecture of the previously released Phi-4, this 14-billion parameter model integrates supervised fine-tuning and reinforcement learning to achieve strong performance on complex problems. According to Microsoft, the Phi-4 reasoning models outperform larger language models on several demanding benchmarks, despite their compact size. This new model pushes the limits of small AI, demonstrating that carefully curated data and training techniques can lead to impressive reasoning capabilities.

The Phi-4 reasoning family, consisting of Phi-4-reasoning, Phi-4-reasoning-plus, and Phi-4-mini-reasoning, is specifically trained to handle complex reasoning tasks in mathematics, scientific domains, and software-related problem solving. Phi-4-reasoning-plus, in particular, extends supervised fine-tuning with outcome-based reinforcement learning, which is targeted for improved performance in high-variance tasks such as competition-level mathematics. All models are designed to enable reasoning capabilities, especially on lower-performance hardware such as mobile devices.

Microsoft CEO Satya Nadella revealed that AI is now contributing to 30% of Microsoft's code. The open weight models were released with transparent training details and evaluation logs, including benchmark design, and are hosted on Hugging Face for reproducibility and public access. The model has been released under a permissive MIT license, enabling its use for broad commercial and enterprise applications, and fine-tuning or distillation, without restriction.

Recommended read:
References :
  • the-decoder.com: Microsoft's Phi-4-reasoning models outperform larger models and run on your laptop or phone
  • MarkTechPost: Microsoft AI Released Phi-4-Reasoning: A 14B Parameter Open-Weight Reasoning Model that Achieves Strong Performance on Complex Reasoning Tasks
  • THE DECODER: Microsoft's Phi-4-reasoning models outperform larger models and run on your laptop or phone
  • AI News | VentureBeat: The release demonstrates that with carefully curated data and training techniques, small models can deliver strong reasoning performance.
  • Maginative: Microsoft’s Phi-4 Reasoning Models Push the Limits of Small AI
  • www.marktechpost.com: Microsoft AI Released Phi-4-Reasoning: A 14B Parameter Open-Weight Reasoning Model that Achieves Strong Performance on Complex Reasoning Tasks
  • www.tomshardware.com: Microsoft's CEO reveals that AI writes up to 30% of its code — some projects may have all of its code written by AI
  • Ken Yeung: Microsoft’s New Phi-4 Variants Show Just How Far Small AI Can Go
  • www.tomsguide.com: Microsoft just unveiled new Phi-4 reasoning AI models — here's why they're a big deal
  • Techzine Global: Microsoft is launching three new advanced small language models as an extension of the Phi series. These models have reasoning capabilities that enable them to analyze and answer complex questions effectively.
  • Analytics Vidhya: Microsoft Launches Two Powerful Phi-4 Reasoning Models
  • www.analyticsvidhya.com: Microsoft Launches Two Powerful Phi-4 Reasoning Models
  • www.windowscentral.com: Microsoft Introduces Phi-4 Reasoning SLM Models — Still "Making Big Leaps in AI" While Its Partnership with OpenAI Frays
  • Towards AI: Phi-4 Reasoning Models
  • the-decoder.com: Microsoft's Phi 4 responds to a simple "Hi" with 56 thoughts
  • Data Phoenix: Microsoft has introduced three new small language models—Phi-4-reasoning, Phi-4-reasoning-plus, and Phi-4-mini-reasoning—that reportedly deliver complex reasoning capabilities comparable to much larger models while maintaining efficiency for deployment across various computing environments.
  • AI News: Phi-4-reasoning, Phi-4-reasoning-plus, and Phi-4-mini-reasoning—that reportedly deliver complex reasoning capabilities comparable to much larger models while maintaining efficiency for deployment across various computing environments.
Aswin Ak@MarkTechPost //

Microsoft Unveils Efficient Phi-4 Multimodal AI Models - Microsoft introduced new Phi-4 AI models (Phi-4-multimodal and Phi-4-mini) for efficient processing of text, speech, and vision data, offering performance improvements with less computing power.
Microsoft has unveiled its new Phi-4 AI models, including Phi-4-multimodal and Phi-4-mini, designed to efficiently process text, images, and speech simultaneously. These small language models (SLMs) represent a breakthrough in AI development, delivering performance comparable to larger AI systems while requiring significantly less computing power. The Phi-4 models address the challenge of processing diverse data types within a single system, offering a unified architecture that eliminates the need for separate, specialized systems.

Phi-4-multimodal, with 5.6 billion parameters, can handle text, speech, and visual inputs concurrently. Phi-4-mini, a smaller model with 3.8 billion parameters, excels in text-based tasks such as reasoning, coding, and instruction following. Microsoft claims Phi-4-mini outperforms similarly sized models and rivals models twice its size on certain tasks. These models aim to empower developers with advanced AI capabilities, offering enterprises cost-effective and efficient solutions for AI applications.

Recommended read:
References :
  • MarkTechPost: Microsoft AI Releases Phi-4-multimodal and Phi-4-mini: The Newest Models in Microsoft’s Phi Family of Small Language Models (SLMs)
  • Analytics Vidhya: Microsoft has officially expanded its Phi-4 series with the introduction of Phi-4-mini-instruct (3.8B) and Phi-4-multimodal (5.6B), complementing the previously released Phi-4 (14B) model known for its advanced reasoning capabilities.
  • venturebeat.com: VentureBeat covers Microsoft's new Phi-4 AI models.
  • THE DECODER: Microsoft expands its SLM lineup with new multimodal and mini Phi-4 models
  • Dataconomy: Microsoft expands Phi line with new multimodal models
  • SiliconANGLE: Microsoft releases new Phi models optimized for multimodal processing, efficiency.
  • MarkTechPost: This article reports on Microsoft's release of Phi-4 AI models, which are designed to be smaller and more efficient while still delivering strong performance. The article highlights the trend in AI development towards creating more compact models.

Benchmarks

Blogs

Research Tools