📊 Anthropic tops OpenAI in paid enterprise adoption 📊
Ramp’s AI Index reveals a historic replacement atop enterprise AI adoption rankings.

For the first time, Anthropic has overtaken OpenAI in paid business adoption, according to fintech firm Ramp’s latest AI Index tracking corporate payments across 50,000+ U.S. businesses. Anthropic rose 3.8% in April to reach 34.4% adoption share, while OpenAI fell 2.9% to 32.3%, even as overall enterprise AI usage climbed to 50.6%. Claude Code anchored much of the swing, helping Anthropic expand beyond technical teams into finance, legal, and research workflows. Over the past year, Anthropic has quadrupled its business adoption while OpenAI grew just 0.3% — a clear sign that competitive pressure is overriding vendor lock-in and that enterprise AI procurement has hit an inflection point.

🧮 DeepMind’s AI Co-Mathematician doubles FrontierMath scores 🧮
DeepMind’s new agentic math system opens researcher access to AI-powered proof strategies.

Google DeepMind published a paper on its AI Co-Mathematician, an agentic system built on Gemini 3.1 designed to help mathematicians tackle unsolved problems. Modeled after AI coding environments like Claude Code, it employs a coordinator agent that breaks research into parallel workstreams, with sub-agents writing code, searching literature, and attempting proofs. An Oxford team resolved an open problem in the Kourovka Notebook after spotting a clever proof strategy inside a rejected output. On Epoch AI’s FrontierMath Tier 4, the system topped the leaderboard at 48%, more than doubling Gemini 3.1 Pro’s raw 19% score — a capability jump that redefines how frontier math research gets done.

🎤 Thinking Machines ships real-time interaction models 🎤
A new architecture built for continuous, low-latency collaboration replaces turn-taking with streaming.

Thinking Machines Lab (TML) unveiled interaction models — a new class of AI purpose-built for continuous, real-time collaboration. Unlike conventional models that rely on turn-taking, TML’s system processes inputs in 200ms chunks and responds in a streaming loop across voice, video, and text simultaneously. A background reasoning model handles slower tasks like searches and tool use, positioning TML as a deliberate counter to the agentic-first trend. The multi-stream architecture is trained from scratch, not fine-tuned — a meaningful departure that could reshape how developers design human-AI interaction going forward and pressure incumbents on latency.

INTERESTING TO KNOW

🤖 Google ships Magic Pointer and AI-native Googlebooks 🤖

Ahead of I/O, Google announced Gemini Intelligence, a cross-device AI platform for Android that enables agentic tasks directly within apps. The headline feature, Magic Pointer, lets users point at any on-screen element — a building, a date, a table — while Gemini interprets context and responds via voice. Google also revealed AI-native Googlebook laptops built with Dell, HP, Lenovo, Acer, and Asus, alongside tools like Create My Widget, Rambler dictation, and Gemini auto-browse in Chrome. This cements AI as Android’s default interface layer.

📱 OpenAI ships Codex to ChatGPT iOS app 📱

OpenAI has rolled out its Codex coding agent to the ChatGPT iOS mobile app, letting developers track and manage long-running AI tasks, review code changes, handle approvals, interact with plugins, and start new tasks — all from their phone. The system connects to ongoing work on laptops, devboxes, or remote environments through a “secure relay layer” that avoids exposing the user’s machine to the open internet while syncing across ChatGPT instances. The launch positions Codex as an always-available development companion and was the headline feature across multiple developer-focused outlets.