Anthropic's AI chatbot Claude is adding more than a million new users daily. Mike Krieger shared the milestone on X. The Instagram co-founder joined Anthropic in 2024, initially as Chief Product Officer, and now leads the AI lab's new experiments division.
Mike Krieger announced that more than a million people are signing up for Claude every day. | Image: Anthropic
The surge likely has several drivers. In the consumer space, Anthropic has benefited from the ongoing Pentagon controversy, where the company is widely considered the moral winner compared to OpenAI.
At the same time, AI adoption is growing across the board. Since October 2025, annualized revenue for both companies is estimated to have roughly doubled—from $13 billion to $25 billion for OpenAI, and from $7 billion to $19 billion for Anthropic. That said, both companies are still burning through cash, with massive costs and liabilities offsetting those revenue figures.
ChatGPT users research products but won't buy there, forcing OpenAI to rethink its commerce strategy
OpenAI wanted to turn ChatGPT into a shopping destination, but only about a dozen retailers signed up and users weren’t buying. Now the company is handing off purchases to app partners like Instacart and Target.
Google is expanding its AI-powered search into the workplace. The company has rolled out the "Canvas" feature in AI mode for all users in the US. Canvas is a workspace built into Google's chat systems that lets users organize projects and plans over time. ChatGPT and Claude offer similar functionality.
The update also adds support for creative writing and coding tasks. Users can create documents or build interactive tools and dashboards directly within search: just type in a prompt, and Google generates a working prototype that pulls together current web information and data from Google's Knowledge Graph. Results can be tested, code viewed, and everything refined through chat. The feature is available now at google.com/ai in the US.
With Canvas, AI mode is looking more and more like the Gemini app. Google seems to be gradually unifying its chat offerings, adding features that blur the line between search and a full AI assistant. At some point, AI mode and Gemini could merge into a single product, giving Google its own direct equivalent to ChatGPT.
Alibaba's lead AI researcher Junyang Lin has unexpectedly resigned. Lin was the driving force behind Alibaba's Qwen model series and shaped the company's open-source strategy. According to Chinese technology portal 36Kr, several core team members followed him out the door, including Binyuan Hui (Qwen coder), Bowen Yu (post-training), and Kaixin Li (Qwen 3.5/VL). A number of younger researchers reportedly also quit on the same day.
via X
Alibaba CEO Eddie Wu accepted Lin's resignation and announced a new "Foundation Model Task Force," which he will lead alongside CTO Zeming Wu and Jingren Zhou, The Information reports. The company says it plans to double down on open source and ramp up AI investments. "In technology, standing still means falling behind," Wu wrote in a memo to the team. "In technology, standing still means falling behind. Advancing foundation models is a core strategic priority for our future."
OpenAI has released its Codex app for Windows. Codex is an AI-powered coding tool that helps developers build software by running multiple agents asynchronously across projects, delegating repeatable tasks through Automations, and connecting agents to tools and workflows via Skills. Developers can review, guide, and step into agent work without losing context.
For the Windows version, OpenAI built its own sandbox that operates at the OS level with restricted tokens, file system access rights, and dedicated sandbox user accounts. This lets AI agents run directly in Windows environments like PowerShell without forcing developers to switch to WSL or virtual machines. OpenAI has published the sandbox code as open source on GitHub.
The Windows app arrives a few weeks after the Mac version, which was downloaded over a million times in its first week, according to OpenAI. More than 500,000 developers had already signed up for the Windows waiting list. OpenAI says Codex now has over 1.6 million weekly active users total. The app is available across all ChatGPT plans.
Meta has signed a multi-year AI licensing deal with News Corp that will pay the Wall Street Journal's parent company up to $50 million annually. The contract runs at least three years and covers content from the US and UK, reports the Wall Street Journal. Meta can use current content for its AI products and tap article archives for training. News Corp already has a deal with OpenAI worth over $250 million over five years. Meta has also signed agreements with CNN, Fox News, and other outlets.
These deals are a mixed bag for the media industry. They bring in revenue for participating publishers during tough times but risk shrinking media diversity by squeezing out those not at the table. They also give platforms more leverage, letting companies like Meta increasingly set the terms. And because contracts are negotiated one by one, they split the publishing landscape in ways that make it harder for the industry to push for better conditions together.
OpenAI's GPT-5.4 could be the leap forward that the just-released 5.3 Instant for ChatGPT wasn't. The model should drop "sooner than you think," according to OpenAI, but no official details have been shared yet.
According to The Information, GPT-5.4 will feature a one-million-token context window, more than double the 400,000 tokens in the current GPT-5.2. That would put OpenAI on par with Google and Anthropic. The model is also expected to be more reliable and make fewer mistakes on longer tasks that can run for several hours, which matters especially for tools like OpenAI's Codex programming agent.
One notable addition is an "extreme" thinking mode that lets the model burn significantly more compute on tough questions. This mode is aimed at researchers rather than everyday users who want quick answers. According to The Information, the more frequent model release cadence is designed to keep expectations in check. The hype around the GPT-5 launch set the bar so high it was nearly impossible to clear, and OpenAI's user growth has recently fallen short of internal projections.
