OpenAI is reportedly planning to launch new AI models, including GPT-4.1 and its mini/nano versions, along with o3 and o4 mini inference models. These could debut next week. CEO Sam Altman previously teased a new feature but details are unclear. Delays have occurred due to capacity issues and resource constraints. OpenAI has yet to confirm these plans officially.

Microsoft is slowing or halting some AI data center projects, including a $1 billion initiative in Ohio, citing reassessment of infrastructure strategy amid global market fluctuations and tariff impacts. The company paused developments in Wisconsin, scaled back expansions, and canceled 2 gigawatt projects due to oversupply concerns. Despite planning an $80 billion investment for AI-enabled data centers in 2025, Microsoft is adjusting its approach, with potential layoffs rumored for mid-2025 to o

Google enhances Vertex AI with a full-stack platform for building, deploying, and managing multi-agent AI systems. Key features include the Agent Development Kit (ADK) for simplified development, the Agent2Agent (A2A) protocol for cross-framework interoperability, and the Agent Engine for managed production deployment. These updates focus on enterprise integration, security, and scalability, enabling businesses to transition from experimental AI to production-ready solutions.

Samsung plans to integrate Google's Gemini AI models into its upcoming Ballie home robot, expanding their AI partnership. Ballie, first previewed in 2020 and set to launch this summer, is a basketball-sized device that navigates homes using wheels. It supports voice commands, projects virtual buttons on the floor, streams videos, greets users, sets reminders, and adjusts smart appliances. With Gemini, it can provide health suggestions via Google searches. Likely using the cost-effective Gemini 2

Google will support Anthropic's Model Context Protocol (MCP) in its Gemini models and SDK, following OpenAI's similar move. MCP, an emerging open standard, enables AI models to access data sources and content libraries, facilitating two-way connections for developers. It is gaining industry adoption since being open-sourced by Anthropic.

Microsoft is testing new Copilot features on Windows, including screen sharing with the AI assistant and expanded Copilot Vision support for all PC applications. Previously limited to Edge, Copilot Vision now guides users in apps like Photoshop and analyzes content. Testing is currently US-limited, with plans to roll out to all Windows 11 users. Additionally, a new file search feature allows AI-assisted searches for various file formats, enhancing productivity. These features work on the standar

Anthropic launches the "Max Plan," a new subscription for power users needing higher usage limits. Priced at $100/month (5x Pro plan allowance) or $200/month (20x allowance), it targets professionals requiring flexible quotas without enterprise deployment. Users gain priority access to new features and capabilities, supporting intensive all-day usage scenarios. This move expands Anthropic's presence in the professional market, competing directly with OpenAI's similar offerings.

Google unveiled new features for Veo 2, enhancing video creation with inpainting, outpainting, and cinematographic techniques. Imagen 3 improves text-to-image editing, while Lyria and Chirp 3 advance audio capabilities with custom voices and speaker separation. Efficiency-boosting Gemini 2.5 Flash and updated Agentic AI tools for cross-platform tasks are also introduced.

Adobe is developing AI agent tools for Photoshop and Premiere Pro to boost creative efficiency. The Photoshop tool, "Creative Agent," features a floating panel offering automated editing suggestions like background removal and subject highlighting, reducing reliance on manual adjustments. It aims to support natural language commands for complex tasks. For Premiere Pro, AI will enable object recognition, rough cut generation, and optimize processes like lens selection and color correction. Adobe

Deep Cogito launched its hybrid AI reasoning models, the Cogito 1 series, available via download and API. These models, ranging from 3B to 70B parameters (with a future 671B version planned), integrate reasoning and non-reasoning modes for efficient problem-solving. Built on Meta’s Llama and Alibaba’s Qwen, they outperform comparable open-source models in benchmarks. Accessible through platforms like Fireworks AI and Together AI, the series aims to advance towards general superintelligence. Foun

Meta introduced Llama 4 models Scout and Maverick, with Maverick outperforming GPT-4o and Gemini 2.0 Flash in benchmarks. It reached second place on LMArena with an ELO rating of 1417. However, the tested version was an experimental chat-optimized variant, not the public release, sparking criticism for lack of transparency. LMArena plans to update its policy for fair evaluations. Meta acknowledged experimenting with customized variants.

Google's AI note-taking tool, NotebookLM, is rumored to launch a mobile app, expanding beyond desktop platforms. Known for extracting insights from documents and offering AI-generated podcasts, the tool will enhance accessibility for on-the-go users. The mobile version aims to boost usability, though details like release date and features are yet unclear. This move could strengthen NotebookLM's market position.

Former OpenAI CRO Bob McGrew joins Thinking Machines Lab as an advisor, alongside other ex-OpenAI researchers. Founded by ex-OpenAI CTO Mira Murati, the company aims to advance AI through open science and practical applications. With 19 former OpenAI employees in its team, Thinking Machines Lab launched in February and reportedly sought $1 billion in funding at a $9 billion valuation.

Amazon unveiled two AI breakthroughs: Nova Sonic for real-time voice interaction with unified architecture and emotion recognition, integrated into Bedrock and Alexa Plus. Nova Reel 1.1 enhances video generation with reduced latency and seamless multi-clip stitching, supporting short videos and ads. These innovations aim to advance voice tech and content creation efficiency in competitive AI markets.

Google introduces Sec-Gemini v1, an experimental AI model for cybersecurity. It accesses real-time data from trusted sources like Google Threat Intelligence and Mandiant to identify threats, analyze incidents, and understand vulnerabilities. Sec-Gemini outperforms other models in security benchmarks, offering rapid insights into root causes and threat actors.

Riverbed Technology updates its observability platform with AI tools like IQ Assist for visual issue resolution, Predictive AI for preemptive problem detection, and Agentic AI for automated troubleshooting. Enhancements include Unified Agent modules for telemetry from Zoom, Microsoft, Wi-Fi, and Thunderbolt, plus NPM+ Packet Capture for network traffic monitoring. Aternity improvements simplify installation in Kubernetes and virtual desktops. Most features are available now, with Agent AI and NP

IBM unveiled the z17 mainframe, powered by the Telum II processor with full encryption and support for over 250 AI use cases. It handles 450 billion daily inference operations, a 50% increase from the z16. Designed with expansion in mind, it initially supports 48 Spyre AI chips, aiming to reach 96 within a year. The system offers high compatibility, energy efficiency, and aligns with AI market trends, set for release on June 8.

Amazon's AI video model, Nova Reel 1.1, now generates two-minute videos with multiple six-second segments, maintaining consistent style. Exclusive to AWS and Bedrock, it offers a "Multi-Shot Manual" mode for user-controlled shot composition. Despite ethical concerns over training data sources and copyright issues, Amazon protects users under its indemnification policy. Nova Reel highlights Amazon's growing focus in generative video amid competition from OpenAI and Google.

Sam Altman and Jony Ive's AI project, io Products, is developing a screenless smartphone-like device and plans to enter the AI-powered home devices market. The project aims to integrate AI for voice, gesture, and sensor interactions, potentially revolutionizing user experiences. OpenAI reportedly considered acquiring io Products for $500 million, showcasing its promising potential in AI hardware innovation. However, all initiatives are still in the exploratory phase.

Google's AI chatbot now supports multimodal capabilities, allowing it to analyze images and provide detailed responses using Gemini AI and Google Lens. This update, employing a "divergent query technique," enhances visual search by interpreting scenes, object relationships, and contextual details in images. Initially exclusive to Google One subscribers, the feature is now expanding to millions of users in the U.S., offering richer AI-driven search experiences akin to rivals like ChatGPT.