Software companies are constantly trying to add more and more AI features to their platforms, and AI companies are constantly releasing new models and features. 

Here are all the major AI updates we covered in the month of March.

Google releases reasoning model Gemini 2.5, its “most intelligent AI model” yet

Gemini 2.0 Flash Thinking was the company’s first reasoning model, and Gemini 2.5 builds on that with a better base model and improved post-training. In its announcement, Google revealed that all of its future AI models will have reasoning capabilities built in.

The first Gemini 2.5 model is Gemini 2.5 Pro Experimental, and it leads in LMArena benchmarks over other reasoning models like OpenAI o3-mini, Claude 3.5 Sonnet, and DeepSeek R1.

“Gemini 2.5 models are thinking models, capable of reasoning through their thoughts before responding, resulting in enhanced performance and improved accuracy. In the field of AI, a system’s capacity for “reasoning” refers to more than just classification and prediction. It refers to its ability to analyze information, draw logical conclusions, incorporate context and nuance, and make informed decisions,” Koray Kavukcuoglu, CTO of Google DeepMind, wrote in a blog post

OpenAI announces 4o Image Generation

The latest image generation model improves on text rendering, has the ability to refine images through multiple follow-up prompts, and offers better instruction following, with the ability to handle up to 10-20 different objects in a prompt. 

It can also perform in-context learning to analyze and learn from user-uploaded images, and the model also links its knowledge between text and images to generate better results. 

4o image generation has begun rolling out for Plus, Pro, Team, and Free users as the default image generator, and access will soon be available for Enterprise and Edu users. 

Microsoft unveils new reasoning agents in Microsoft 365 Copilot

The two agents, Researcher and Analyst, can help users analyze vast amounts of information, spanning emails, meetings, files, chats, and more. 

Researcher is ideal for multi-step research, such as building a go-to-market strategy based on both the context of a company’s work and broader competitive data found online. Beyond data in Microsoft 365, it can also leverage third-party connectors to bring in data from sources like Salesforce, ServiceNow, and Confluence.

Analyst is designed for complex data analysis, such as turning raw data from multiple spreadsheets into a demand forecast for a new product or a visualization of customer purchasing patterns. 

These two agents will begin rolling out to Microsoft 365 Copilot subscribers starting in April as part of the Frontier early access program. 

Microsoft Security Copilot gets several new agents

The new agents include a Phishing Triage Agent in Microsoft Defender, Alert Triage Agents in Microsoft Purview, Conditional Access Optimization Agent in Microsoft Entra, Vulnerability Remediation Agent in Microsoft Intune, and Threat Intelligence Briefing Agent in Security Copilot.

The company also announced five additional agents from its Microsoft Security partners: Privacy Breach Response Agent by OneTrust, Network Supervisor Agent by Aviatrix, SecOps Tooling Agent by BlueVoyant, Alert Triage Agent by Tanium, and Task Optimizer Agent by Fletch.

The agents will be available in preview starting in April.

“Building on the transformative capabilities of Security Copilot, the six Microsoft Security Copilot agents enable teams to autonomously handle high-volume security and IT tasks while seamlessly integrating with Microsoft Security solutions. Purpose-built for security, agents learn from feedback, adapt to workflows, and operate securely—aligned to Microsoft’s Zero Trust framework. With security teams fully in control, agents accelerate responses, prioritize risks, and drive efficiency to enable proactive protection and strengthen an organization’s security posture,” Vasu Jakkal, corporate vice president of Microsoft Security, wrote in a blog post.

Red Hat AI offers new capabilities across Red Hat OpenShift AI

Red Hat OpenShift AI 2.18 adds new features such as distributed serving that allows IT teams to split model serving across multiple GPUs, an end-to-end model tuning experience across InstructLab and Red Hat OpenShift AI data science pipelines, and model evaluation.

This release also includes a preview of AI Guardrails, which offer additional methods for identifying and mitigating “potentially hateful, abusive or profane speech, personally identifiable information, competitive information or other data limited by corporate policies.”

Akamai launches new platform for AI inference at the edge

Akamai has announced the launch of Akamai Cloud Inference, a new solution that provides tools for developers to build and run AI applications at the edge.

According to Akamai, bringing data workloads closer to end users with this tool can result in 3x better throughput and reduce latency up to 2.5x.

Akamai Cloud Inference offers a variety of compute types, from classic CPUs to GPUs to tailored ASIC VPUs. It offers integrations with Nvidia’s AI ecosystem, leveraging technologies such as Triton, TAO Toolkit, TensorRT, and NVFlare.

Due to a partnership with VAST Data, the solution provides access to real-time data so that developers can accelerate inference-related tasks. The solution also offers highly scalable object storage and integration with vector database vendors like Aiven and Milvus.

AlexNet source code is now open source

AlexNet is a neural network for recognizing images that was created in 2012 by University of Toronto graduate students Alex Krizhevsky and Ilya Sutskever and their advisor Geoffrey Hinton.

“Before AlexNet, almost none of the leading computer vision papers used neural nets. After it, almost all of them would. AlexNet was just the beginning. In the next decade, neural networks would advance to synthesize believable human voices, beat champion Go players, model human language, and generate artwork, culminating with the release of ChatGPT in 2022 by OpenAI, a company cofounded by Sutskever,” wrote Hansem Hsu, curator of the Computer History Museum Software History Center, the organization that is releasing the source code, in partnership with Google.  

The source code can be found here

Anthropic’s Claude can now search the web when generating responses

Anthropic has announced that Claude can now search the Internet, allowing it to generate more up-to-date and relevant responses.

For instance, a developer who is getting an error updating a dependency in TypeScript 5.5 could ask Claude if there were any breaking changes between version 5.4 and 5.5 and also ask for recommended fixes.

Claude will respond with direct citations of its web sources, allowing users to fact check the information.

Google launches Canvas to enable easier collaboration with Gemini

Google is making it easier for developers to collaborate with Gemini with the launch of Canvas, an interactive space to create and refine code. 

Canvas could be used to build reports, blog posts, study guides, visual timelines, interactive prototypes, code snippets, and more. 

The new tool makes it easier for users to refine their work, such as highlighting a paragraph and asking Gemini to make it more concise or professional.

OpenAI adds new audio models to API

The new speech-to-text and text-to-speech models will enable developers to “build more powerful, customizable, and intelligent voice agents that offer real value,” according to OpenAI.  

The updated speech-to-text models perform particularly well in scenarios involving accents, noisy environments, and fluctuating speech speeds, improving transcription quality. This makes them particularly well-suited for use cases such as call centers and meeting note transcription, OpenAI explained.

Developers will now be able to prompt the text-to-speech model to speak in a certain way, such as “talk like a sympathetic customer service agent.”

Nvidia unveils several AI advancements at GTC

During its GTC conference this week Nvidia made a number of announcements related to AI, including AI-Q Blueprint, which is a system for building agentic systems. It provides references for integrating Nvidia accelerated computing, partner storage platforms, and software and tools. 

The company also announced a family of open reasoning AI models, Llama Nemotron, which are based on Meta’s Llama models and offer improvements over the base model in multistep math, coding, reasoning, and complex decision making. 

A full list of announcements from GTC can be found here

IBM Research announces Agent Communication Protocol

Agent Communication Protocol (ACP) is a standard for agent communication to enable interoperability, simplified development, and the ability to reuse solutions. 

ACP is an extension of Model Communication Protocol (MCP), which is a standard introduced by Anthropic to standardize how apps and LLMs communicate. 

“Current agent systems often use diverse communication standards, causing complexity, integration difficulties, and vendor lock-in. ACP addresses these issues uniquely by standardizing interactions tailored specifically for agents that handle natural language inputs and depend on externally hosted models. By accommodating these agent-specific needs, ACP simplifies integration and promotes effective collaboration across agent-based ecosystems,” the draft proposal states. 

Oracle launches AI Agent Studio

AI Agent Studio is a platform for creating, extending, deploying, and managing AI agents and agent teams. It is part of Oracle Fusion Cloud Applications Suite, and includes over 50 pre-packaged agents.

It offers capabilities like agent template libraries, agent team orchestration, extensibility of the prepackaged agents, flexibility in LLM choice, third-party system integration, a trust and security framework, and validation and testing tools.

“AI agents are the next phase of evolution in enterprise applications and just like with existing applications, business leaders need the flexibility to create specific functionality to address their unique and evolving business needs,” said Steve Miranda, executive vice president of applications at Oracle. “Our AI Agent Studio builds on the 50+ AI agents we have already introduced and gives our customers and partners the flexibility to easily create and manage their own AI agents. With the agents already embedded in Fusion Applications and our new AI Agent Studio, customers will be able to further extend automation and ultimately, achieve more while spending less.”

WSO2 updates AI-powered IDP Choreo

The latest release adds new capabilities such as:

  • Customizable CI pipelines and parallel deployment options
  • AI-driven cost insights, including recommendations for how to optimize costs
  • Automatic alerts from metrics and logs
  • Support for local pipelines and observability

Choreo’s AI copilot has also been updated with support for encryption keys for APIs, hotfix deployment pipelines, and support for environment-aware configuration groups and unified configuration declaration.

And finally, WSO2 is also releasing an open source version of Choreo. 

“AI holds an opportunity for enterprises seeking to compete with new intelligent digital experiences, but the complexity of today’s infrastructure is hindering their efforts,” said Kanchana Wickremasinghe, WSO2 vice president and general manager of Choreo at WSO2. “The latest release of our Choreo AI-native IDP, available in the cloud and as open-source software, is clearing the way for enterprises to innovate by extending AI capabilities that help software engineers deliver new apps faster while enabling platform engineers to quickly respond to developers’ ever-changing requirements and expectations.”

Stravito enhances its generative AI assistant with new capabilities

Stravito Assistant now has a Focus Mode where it will go into a deep analysis mode when given a set of reports, videos, or collections to detect patterns and insights from those collections of data.

Another new feature—Snapshots—provides instant summaries of a report, so that users can get the key takeaways from a document quickly. And additionally, Stravito Assistant now supports over 100 languages.

“These updates reinforce our commitment to providing purpose-built AI-powered tools that help global enterprises leverage their market research to make data-driven, cost-effective decisions that fuel innovation and long-term growth,” said Thor Olof Philogène, founder and CEO of Stravito.

Google announces Gemma 3

Gemma 3 is Google’s latest AI model, offering improved math, reasoning, and chat capabilities. It can handle context windows of up to 128k tokens, understand 140 languages, and comes in four sizes: 1B, 4B, 12B, and 27B.

It is a multimodal model, and it supports images and videos as inputs, which allows it to analyze images, answer questions about a picture, compare images, identify objects, or reply about text on an image. 

Gemma 3 is available as either a pre-trained model that can be fine-tuned for specific use cases, or as a general-purpose instruction-tuned model. It is available in Google AI Studio, or can be downloaded through Hugging Face or Kaggle.

OpenAI reveals Responses API, Agents SDK for building agentic experiences

OpenAI is releasing new tools and APIs to help developers build agentic experiences. The Responses API allows developers to more easily integrate OpenAI’s tools into their own applications. 

“As model capabilities continue to evolve, we believe the Responses API will provide a more flexible foundation for developers building agentic applications. With a single Responses API call, developers will be able to solve increasingly complex tasks using multiple tools and model turns,” OpenAI wrote. 

The Responses API comes with several built-in tools, including:

  • Web search, which allows for retrieval of information from the Internet
  • File search, which allows for retrieval of information from large volumes of documents
  • Computer use, which captures mouse and keyboard actions generated by a model so that developers can automate computer tasks.

OpenAI also announced the Agents SDK, an open source tool for orchestrating multi-agent workflows. According to OpenAI, the Agents SDK can be used for a variety of scenarios, including customer support automation, multi-step research, content generation, code review, and sales prospecting.

Boomi launches AI Studio

Boomi AI Studio is a platform for designing, governing, and orchestrating AI agents at scale. It consists of multiple components, including:

  • Agent Designer, which provides no-code templates for building and deploying agents
  • Agent Control Tower, which provides monitoring of agents
  • Agent Garden, which allows developers to interact with agents in natural language
  • Agent Marketplace, where developers can find and download AI agents from Boomi and its partners.

“With Boomi AI Studio, we’re giving organizations a powerful yet accessible way to build, monitor, and orchestrate AI agents with trust, security, and governance at the core,” said Ed Macosky, chief product and technology officer at Boomi. “As of today, Boomi has deployed more than 25,000 AI Agents for customers. This strong market adoption of our AI agents highlights not only the real value they’re delivering, but also the need for a solution that enables organizations to leverage AI responsibly while accelerating innovation and achieving transformative outcomes.”

Amazon SageMaker Unified Studio is now generally available

The platform allows developers to find and access all of the data in their organization and act on it using a variety of AWS tools, such as Amazon Athena, Amazon EMR, AWS Glue, Amazon Redshift, Amazon Managed Workflows for Apache Airflow (Amazon MWAA), and SageMaker Studio.

It was first announced as a preview at AWS re:Invent last year, and new capabilities added since then include support in Amazon Bedrock for foundation models like Anthropic Claude 3.7 Sonnet and DeepSeek-R1, and integration with the generative AI assistant Amazon Q developer.  

Amazon SageMaker Unified Studio is available in US East (N. Virginia, Ohio), US West (Oregon), Asia Pacific (Seoul, Singapore, Sydney, Tokyo), Canada (Central), Europe (Frankfurt, Ireland, London), and South America (São Paulo) AWS regions. 

“SageMaker Unified Studio breaks down silos in data and tools, giving data engineers, data scientists, data analysts, ML developers and other data practitioners a single development experience. This saves development time and simplifies access control management so data practitioners can focus on what really matters to them—building data products and AI applications,” Donnie Prakoso, principal developer advocate at AWS, wrote in a blog post. 

Visual Studio now includes access to GPT-4o Copilot code completion model 

The code completion model was trained on over 275,000 public repositories in 30 different programming languages, on top of the GPT-4o training. This results in more accurate completion suggestions, Microsoft explained. 

It will be available for users working in Visual Studio 17.14 Preview 2, which was released this week.  

SUSE AI is updated with new features for agentic AI use cases

SUSE AI is an open infrastructure platform for running AI workloads, and the latest release includes a number of new features, such as:

  • Tools and blueprints for developing agentic workflows
  • New observability features that provide insights into LLM token usage, GPU utilization, performance bottlenecks, and more
  • LLM guardrails to ensure ethical AI practices, data privacy, and regulatory compliance
  • SUSE AI Library now includes support for OpenWebUI Pipelines and PyTorch 

“Through close collaboration with our customers and partners since the launch of SUSE AI last year, we’ve gained additional and invaluable insights into the challenges of deploying production-ready AI workloads,” said Abhinav Puri, general manager of Portfolio Solutions & Services at SUSE. “This collaborative journey has allowed us to bolster our offerings and continue to provide customers strong transparency, trust, and openness in AI implementation. These new enhancements reflect our commitment to building on that partnership and delivering even greater value, while strengthening SUSE AI.”

Eclipse Foundation releases Theia AI

Theia AI is an open source framework for integrating LLMs into tools and IDEs. It gives developers full control and flexibility over how AI is implemented into their applications, from orchestrating the prompt engineering flow to defining agentic behavior to deciding which data sources are used.

Additionally, the organization said that an AI-powered Theia IDE based on the Theia AI framework is now in alpha. The Eclipse Foundation says this IDE will give developers access to AI-enhanced development tools while also allowing them to maintain user control and transparency. 

Both tools are being contributed to the Eclipse Foundation by EclipseSource. “We believe that openness, flexibility, and transparency are key success factors for the innovative and sustainable adoption of AI in tools and IDEs,” said Jonas Helming, CEO of EclipseSource. “Large language models inherently introduce a significant level of indeterminism into modern workflows. Developers don’t need yet another proprietary black-box layer they cannot control and adapt. For tool builders developing reliable industrial solutions, it is even more crucial to have full customizability and control over every aspect of an AI-powered tool while also benefiting from a robust framework that allows them to focus on their domain-specific optimizations.” 

Anthropic makes changes to reduce token usage

The company announced several new features to help users spend fewer tokens when interacting with its models:

  • Cache-aware rate limits: Prompt cache read tokens don’t count against the Input Tokens Per Minute (ITPM) limit anymore on Claude 3.7 Sonnet, allowing users to optimize their prompt caching to get the most of their ITPM limit.
  • Simpler prompt caching management: When a cache breakpoint is set, Claude will now automatically read from the longest previously cached prefix. This means that users won’t have to manually track and specify which cached segment to use, as Claude will automatically identify the most relevant one. 
  • Token-efficient tool use: Users can now specify that Claude calls tools in a token-efficient manner, resulting in up to a 70% reduction on output token consumption (the average reduction has been 14% among early adopters).

Diffblue releases tool for verifying its AI-generated unit tests

Diffblue Test Review was designed to give developers more confidence in accepting AI-generated unit tests. A recent Stack Overflow study found that only 2% of developers have confidence that AI-generated code is accurate. Test Review aims to give developers the insights needed to make an informed decision about accepting tests into their codebase. 

Developers can review each test and accept them all in one click, or send specific tests back or edit them before accepting them into the codebase. 

“We hope to win over developers who are apprehensive about integrating a fully-autonomous agent into their development workflow,” said Peter Schrammel, co-founder and CTO of Diffblue. “By lowering the barrier to adoption, developers can ease into an AI-powered iterative unit testing workflow, and ultimately, evolve into full autonomy and the remarkable scalability that results from it.”

ScaleOut Software adds generative AI to Digital Twins service

ScaleOut Digital Twins provides a framework for building and running digital twins at scale. Version 4 adds capabilities such as automatic anomaly detection using AI, the ability to use natural language prompts to create data visualizations, the ability to retrain machine learning algorithms in live systems, and other performance improvements.  

“ScaleOut Digital Twins Version 4 marks a pivotal step in harnessing AI and machine learning for real-time operational intelligence,” said Dr. William Bain, CEO and founder of ScaleOut Software. “By integrating these technologies, we’re transforming how organizations monitor and respond to complex system dynamics — making it faster and easier to uncover insights that would otherwise go unnoticed. This release is about more than just new features; it’s about redefining what’s possible in large-scale, real-time monitoring and predictive modeling.”

JFrog launches end-to-end DevSecOps platform for deploying AI applications

JFrog is releasing a new end-to-end solution for developing and deploying enterprise AI applications that brings together development teams, data scientists, and machine learning engineers into a single platform. 

JFrog ML provides a holistic view of the entire AI software supply chain, from software packages to LLMs, so that companies can ensure their AI applications are secured in the same way their traditional software is. 

It provides security scanning for AI models, whether they were created in-house or are from a third-party.

Other key features include providing a single system of record, reproducible artifacts for all models created in the platform, simplified model development and deployment processes, and dataset management and feature store support.

Anthropic Console now facilitates prompt collaboration

Developers will now be able to share prompts with others through the Console. Team members have access to a shared library of prompts, eliminating the need to copy and paste prompts to share them. 

Additionally, Anthropic Console now supports the company’s latest model, Claude 3.7 Sonnet, and offers new capabilities to assist users in writing prompts for that model’s extended thinking mode, as well as setting the budget for extended thinking. 

Salesforce launches Agentforce 2dx

Agentforce is the company’s platform for integrating AI agents into employee workflows, and Agentforce 2dx introduces new features that make it even easier for AI agents to be set up. 

Capabilities include a new API, the ability to embed Agentforce into Salesforce business logic, new integrations with MuleSoft, integration with the Slack Workflow Builder, new employee templates for Agentforce use cases, and more. Certain features have already begun rolling out, and Agentforce 2dx is expected to be fully available in April. 

“By extending digital labor beyond CRM, we’re making it easier than ever for businesses to embed agentic AI into any workflow or application to handle routine tasks, augment employees, and connect with customers,” said Adam Evans, EVP and GM of Salesforce’s AI Platform. “With deep integrations across Salesforce’s digital labor platform, CIOs, IT leaders, and developers can seamlessly build agents and automate work wherever it happens, driving efficiency, fueling innovation, and unlocking new opportunities in the $6 trillion digital labor market.”

Sonatype announces AI Software Composition Analysis

This end-to-end tool allows companies to protect and manage their models throughout development and deployment. 

It blocks malicious models from entering development environments, provides a centralized method for governance, automates policy management, and offers full visibility into model consumption.

“No one knows open source like Sonatype, and AI is the next frontier. Just as we revolutionized open source security, we are now doing the same for AI,” said Mitchell Johnson, chief product development officer at Sonatype.

Moderne launches AI agent for code refactoring

Moderne is the creator of the open-source project, OpenRewrite, which automates mass code refactorings. The new AI agent, Moddy, has access to OpenRewrite’s capabilities, enabling developers to navigate, analyze, and modify large, multi-repository codebases.

For instance, a developer could ask Moddy to describe the dependencies that are in use, upgrade frameworks, fix vulnerabilities, or locate specific business logic. 

Its Lossless Semantic Tree (LST) data model allows it to understand the structure, dependencies, and relationships across multiple repositories. 

“Moddy, the new multi-repo AI agent from Moderne, represents a paradigm shift in how enterprise codebases are managed, maintained, and modernized. It empowers developers to take command of their entire codebase—not just the code in their IDE,” Moderne wrote in a blog post

Google expands AI Overviews, adds AI Mode to Search

The AI Overview feature now utilizes Gemini 2.0, allowing it to answer harder questions, such as those related to coding, math, or multimodal queries. 

AI Mode extends AI Overviews further by allowing users to ask follow-up questions when they get their response, rather than having to start multiple searches to get the information they’re looking for. 

For instance, a user could ask “what’s the difference in sleep tracking features between a smart ring, smartwatch and tracking mat,” and then ask a follow-up question: “what happens to your heart rate during deep sleep?”

Amazon Bedrock Data Automation is now generally available

First announced in preview during AWS re:Invent last year, Amazon Bedrock Data Automation streamlines the process of getting insights from unstructured, multimodal content, like documents, images, audio, and videos. 

“With Bedrock Data Automation, you can reduce the development time and effort to build intelligent document processing, media analysis, and other multimodal data-centric automation solutions,” the company wrote in a post

Currently, this feature is available in US East (N. Virginia) and US West (Oregon), and AWS plans to expand it to more regions in Europe and Asia later this year. 

Microsoft open sources Microsoft.Extensions.AI.Evalutions library

This library provides a framework for evaluating the quality of AI applications, and it is now available as part of the dotnet/Extensions repository, which contains a number of libraries useful in creating production-ready applications. 

Along with the open source release, Microsoft is also providing a new set of samples to help developers get started with the library. The samples showcase common use cases and demonstrate how to leverage the library’s capabilities. 

OpenAI announces consortium for using AI to advance research and education

NextGenAI is a collaboration between OpenAI and 15 research institutions to use AI to “accelerate research breakthroughs and transform education.”

The participating institutions include Caltech, the California State University system, Duke University, the University of Georgia, Harvard University, Howard University, Massachusetts Institute of Technology, the University of Michigan, the University of Mississippi, The Ohio State University, the University of Oxford, Sciences Po, Texas A&M University, Boston Children’s Hospital, and the Boston Public Library.

OpenAI is committing $50 million in research grants, compute funding, and API access to those organizations. 

“The field of AI wouldn’t be where it is today without decades of work in the academic community. Continued collaboration is essential to build AI that benefits everyone. NextGenAI will accelerate research progress and catalyze a new generation of institutions equipped to harness the transformative power of AI,” said Brad Lightcap, COO of OpenAI.

Teradata launches new solution for efficiently handling vector data for agentic AI use cases

Teradata, a provider of data analytics solutions, announced a new database offering for managing vector data.

Teradata Enterprise Vector Store manages unstructured data in multi-modal formats like text, video, images, and PDFs. It can process billions of vectors and integrate them into pre-existing systems, and offers response times in the tens of milliseconds. 

According to the company, vector stores are an important foundation for agentic AI, but many vector stores require organizations to make tradeoffs, such as getting fast results, but only for small data sets, or being able to handle large vector volumes, but not at the speed required by agentic AI use cases.