Good morning. It’s Wednesday, May 7th.

On this day in tech history: 2014: Scientists created the first realistic "virtual universe," simulating 13 billion years of cosmic evolution in a cube with 350 million light-year-long sides and unprecedented resolution.

In today’s email:

• Google’s Gemini 2.5 Pro

• OpenAI’s New Playbook on Nonprofit Control

• Lightricks Launches Open-Source AI Video Model

• 4 New AI Tools

• Latest AI Research Papers

Google's Gemini 2.5 Pro Leads AI Coding, Now With Video-to-App Generation

Google's Gemini 2.5 Pro I/O Edition has emerged as the leader in AI coding, surpassing Claude 3.7 Sonnet with a top score of 1499.95 on the WebDev Arena leaderboard. This version enhances web app generation, UI design, and developer workflows, thanks to improved error reduction and function calling. It supports video content analysis, allowing users to generate applications directly from video input.

The update also facilitates creating code-based visualizations from images and automates web project feature integration, such as CSS styling for elements like color and spacing. Developers can now use Gemini 2.5 Pro for multimodal tasks, combining text, images, video, and code to streamline complex programming tasks. For example, it enables the creation of interactive learning platforms from videos, like converting YouTube content into learning apps. Available via Gemini API and Vertex AI, the update promises greater reliability, fewer errors, and no price increase over the previous version.

Additionally, Google’s new "Simplify" feature for iOS allows users to convert complex text into simpler, digestible content, enhancing user understanding particularly for challenging topics like science and technology without losing key details. This feature is rolling out to iOS users this week. Read more.

OpenAI’s New Playbook on Nonprofit Control, Windsurf Buyout and Microsoft Standoff

OpenAI has reversed its decision to transition into a for-profit entity, confirming that its nonprofit will retain control over the company's operations moving forward. As part of this strategic pivot, it will convert its for-profit arm into a Delaware public benefit corporation under its nonprofit parent, replacing the capped-profit model with standard shares, while assuring regulators and AI-safety advocates that control remains unchanged.

Financially, OpenAI plans to cut the ~20% revenue share it pays Microsoft—its $13.75B investor and exclusive Azure partner—to 10% across all partners by 2030. While SoftBank’s $40B investment talks remain on track, Microsoft has yet to endorse the new structure, seeking guarantees that its ownership stake will not be diluted. The company is also acquiring coding start-up Windsurf for ~$3B, its largest buyout to date, strengthening its developer tools against rivals like GitHub Copilot and Claude Code.

Although a philanthropic advisory board is planned, observers question whether it will hold real sway or merely echo the symbolic oversight seen at other tech giants. Legal experts also note that retaining core intellectual property under the nonprofit may complicate a future IPO, potentially limiting investor influence. The latest restructure trades one set of governance complexities with another, leaving OpenAI’s effort to balance mission and margin to be tested over time. Read more.

Lightricks Launches Open-Source AI Video Model with 30x Speed

Lightricks has launched LTX Video 13B, an open-source AI video model that builds on its 2B parameter predecessor, offering enhanced scale and precision. Running efficiently on consumer GPUs, it delivers high-resolution video up to 30 times faster than similar models. A new multi-scale rendering feature allows creators to gradually enhance video details, while upgraded controls enable camera motion, keyframe editing, multi-shot sequencing, and scene-level adjustments.

Trained on licensed Getty and Shutterstock data, LTX Video 13B ensures commercial-safe usage with high visual quality. Available on Hugging Face and GitHub, the model is free for organizations earning under $10M, allowing developers to fine-tune and integrate it. Read more.

• Hugging Face Unveils $100 3D-Printed Robotic Arm and Free AI Agent Tool

• Anthropic launches "AI for Science" grant program

• HeyGen Drops Avatar IV, AI for Lifelike Animated Videos

• Ace Studio Drops Open-Source, Lightning-Fast Music AI

• Watch: David Sacks Explains How AI Will Go 1,000,000x in Four Years

• A stealth AI model beat DALL-E and Midjourney on a popular benchmark — its creator just landed $30M

• Former Google CEO Eric Schmidt-backed FutureHouse previews an AI tool for 'data-driven' biology discovery

• ServiceNow and Nvidia's new reasoning AI model raises the bar for enterprise AI agents

• Reddit will tighten verification to keep out human-like AI bots

• Famed AI researcher launches controversial startup to replace all human workers everywhere

• Oura adds AI-powered glucose tracking and meal logging

• Nvidia launches fully open source transcription AI model Parakeet-TDT-0.6B-V2 on Hugging Face

• Musk's xAI joins TWG Global, Palantir for AI push in financial sector

• New AI Agent WebThinker Enhances Large Reasoning Models (LRM) Web Reasoning

• Meta, Cisco put open-source LLMs at the core of next-gen SOC workflows

• Over 250 CEOs sign open letter supporting K-12 AI and computer science education

• ByteDance drops UI-TARS-1.5, topping OpenAI’s Operator with 42.5% on OSWORLD

• Meta’s AI app is a nightmarish social feed

• Robots get street smarts as Unitree taps Reborn’s motion‑data engine

• Anduril is working on the difficult AI-related task of real-time edge computing

• AI to flag suspicious behavior on New York City subways

• World's first Amazon-style weapons marketplace debuts with drones, robots, ammo

📄 Llama-Nemotron: Efficient Reasoning Models

📄 RM-R1: Reward Modeling as Reasoning ​​

📄 Voila: Voice-Language Foundation Models for Real-Time Autonomous Interaction and Voice Role-Play

📄 Unified Multimodal Chain-of-Thought Reward Model through Reinforcement Fine-Tuning 

📄 Absolute Zero: Reinforced Self-play Reasoning with Zero Data

Thank you for reading today’s edition.

Your feedback is valuable. Respond to this email and tell us how you think we could add more value to this newsletter.

Interested in reaching smart readers like you? To become an AI Breakfast sponsor, reply to this email or DM us on 𝕏!