PLUS: An AI co-developer's Linux desktop, Stanford's robot designer, and Anthropic cracks post-quantum crypto
Happy reading
OpenAI just announced that its revenue from July alone has already outpaced the total for the entire second quarter. This massive growth appears to be fueled by the recent launch of its powerful new GPT-5.6 model family.
The company's strategy of pairing new models with practical enterprise tools is clearly paying off. Is this aggressive commercialization creating an insurmountable lead, or can competitors like Anthropic close the gap?
In today’s Next in AI:
OpenAI’s record July revenue
An AI co-developer's Linux desktop
Stanford's model that designs custom robots
Anthropic cracks post-quantum crypto
OpenAI's Record-Breaking Month
Next in AI: Fresh off the launch of its new GPT-5.6 models, OpenAI’s CFO revealed the company's annualized recurring revenue in July alone has already surpassed the total for the entire second quarter, signaling massive customer adoption.
Explained:
The growth is powered by the new GPT-5.6 family of models, which introduces a flagship model named Sol, a balanced version called Terra, and a cost-effective option, Luna, all designed to deliver more capability per dollar.
This momentum shows OpenAI is regaining ground against competitors like Anthropic, with company leaders noting that some customers are switching back from rivals after receiving unexpectedly high bills.
The financial success stems from a strong enterprise push, with rapid adoption of the ChatGPT Work platform and the company’s powerful coding assistant, Codex, helping to justify its significant infrastructure investments.
Why It Matters: OpenAI's financial success validates its strategy of pairing powerful new models directly with practical, high-value enterprise tools. This aggressive commercialization is setting a new pace for the industry, accelerating how quickly businesses can integrate AI into their core operations.
The AI Co-Developer
Next in AI: In just six months, a single developer leveraged AI to build Starling, a fully functional Linux desktop environment from the ground up. The project highlights a dramatic reduction in the labor required to build large-scale software.
Explained:
Unlike previous concepts, Starling is a complete desktop session that directly drives the GPU, runs apps it didn't write—like Chrome and Zoom—and natively supports both Wayland and X11.
A solo developer directed the AI to produce over 335K lines of Swift, C, and C++ code, creating the desktop shell, rendering engine, and a full suite of first-party apps in about six months.
This massive reduction in development cost signals a new era for ambitious software, empowering individuals to tackle challenges that previously required large institutional efforts.
Why It Matters: Starling serves as powerful proof that AI is evolving from a task automator into a true development partner for complex systems. This shift empowers individual creators to build software at a scale and speed that was previously unimaginable.
The Shape Of Things To Come
Next in AI: Stanford researchers unveiled a new model that designs complete, optimized robots based on human demonstrations. Called “Transformer Transformer,” it generates a robot’s entire embodiment and control system to master a specific task.
Explained:
The system uses a novel representation called RoboTokens to describe a robot’s entire structure (links, joints, motors) and dynamics (states, actions) in a single, unified format.
Instead of a complex pipeline, a single diffusion transformer model plays three roles: it generates the robot design, validates its performance, and directly controls the finished robot.
In a real-world test, a custom-designed robot for flinging cloth reduced tracking error by 73% and max joint speed by 30% compared to the original, showing the practical benefits of task-specific design.
Why It Matters: This marks a fundamental shift from designing general-purpose robots to dynamically generating specialized hardware tailored for any given task. This approach could rapidly accelerate the development of custom robots for everything from manufacturing lines to household chores.
Anthropic's AI Cryptanalyst
Next in AI: Anthropic’s advanced AI model discovered a significant weakness in a proposed post-quantum signature scheme. The breakthrough shows AI's growing capability to perform complex cryptanalysis, a task traditionally reserved for human experts.
Explained:
The AI-driven attack targeted HAWK, a candidate for a future post-quantum standard, effectively halving its security and likely ending its path to standardization.
A second finding improved an attack on a weakened 7-round version of AES, but this is an incremental advance and poses no threat to the full AES encryption protecting data today.
The AI didn't invent new mathematics but instead showcased a powerful ability to apply existing tools more thoroughly, a finding an expert analyst described as "a little embarrassing for the field."
Why It Matters: This demonstrates that advanced AI can serve as a powerful audit tool, finding critical flaws in proposed security standards before they are widely adopted. It signals a shift where AI-human collaboration will become essential for building and stress-testing the next generation of digital infrastructure.
AI Pulse
Theo-Conjecture proved a 35-year-old math problem first posed by an early AI program called Graffiti, discovering an unexpected second-order term that human mathematicians had not predicted.
Linus affirmed his support for using AI tools in Linux kernel development, stating the project will continue to be about technology and rejecting ethical arguments against their use.
Samsung posted a record quarterly operating profit that was up 1,814% year-over-year, driven by soaring demand for its memory chips used in AI servers.
OpenAI revealed that enabling two API settings—retained reasoning and compaction—tripled its GPT-5.6 Sol model’s score on the ARC-AGI-3 benchmark while cutting output tokens by 6x.