OpenAI's Astra Cracks a Decade of Unsolved Mathematics
In what may be the most dramatic demonstration of AI-driven scientific capability to date, OpenAI revealed that an internal version of its next major model — codenamed Astra — has produced solutions to ten previously unsolved problems in mathematics and theoretical computer science. Each problem had remained open for at least a decade.
The headline result is the first-ever explicit construction of a non-sofic group, resolving a central question in group theory that has stood since Mikhail Gromov introduced the concept of soficity in 1999. Other results include disproving Connes's rigidity conjecture on von Neumann algebras, proving Ehrhart's volume conjecture, and contributions spanning high-dimensional geometry, coding theory, quantum complexity, lattice cryptography, and extremal combinatorics.
OpenAI published a 249-page manuscript alongside machine-checkable Lean 4 certificates for every result on GitHub, reporting a "sorry" count of zero — meaning no step in any proof was left unverified by the proof assistant. The entire compute cost was roughly $2,000.
The results have not yet undergone formal peer review, and OpenAI has not announced a public release date for Astra, describing it only as its "next major model." Still, the achievement has sent shockwaves through both the AI and mathematics communities.
Sources: OpenAI, The Next Web, SiliconANGLE
Anthropic Posts First-Ever Operating Profit on $11.5 Billion Q2 Revenue
Anthropic has crossed a historic milestone: the Claude maker reported preliminary Q2 2026 revenue exceeding $11.5 billion — a more than 14-fold increase from the same period last year — and projected its first quarterly operating profit of $559 million. The company had previously told investors it wouldn't reach profitability before 2028, making this achievement roughly two years ahead of schedule.
The numbers represent a dramatic acceleration from Q1's $4.73 billion in revenue, suggesting that enterprise adoption of Claude is scaling rapidly. According to separate reporting, Anthropic has also surpassed OpenAI in quarterly revenue for the first time, a symbolic shift in the competitive landscape of frontier AI labs.
On the leadership front, former Stanford law professor Mariano-Florentino (Tino) Cuellar joined Anthropic as Chief Global Affairs Officer on August 4, signaling the company's growing focus on policy and regulatory engagement as AI governance becomes increasingly central to the industry.
Meta Releases Muse Glimmer: A 30B Open-Weight Model You Can Run on Your Laptop
Meta Superintelligence Labs released Muse Glimmer, a 30-billion-parameter model optimized for on-device agentic workflows, under a fully permissive Apache 2.0 license. The model is small enough to run on a Mac or PC with a single consumer GPU, opening up use cases from local coding agents to LLM-as-a-judge evaluation — all without sending data to the cloud.
Technically, Glimmer is a dense causal transformer with approximately 29.6 billion parameters across 52 layers, including a dedicated ~1.8B-parameter ViT-G/14 perception encoder. It accepts interleaved text and images, supports over 100 languages, and offers a context length of 131,072 tokens with a knowledge cutoff of January 2026.
The release is part of Meta's broader push to challenge OpenAI and Anthropic through open-source AI. CEO Mark Zuckerberg recently published a manifesto titled "The future is for everyone," arguing for the release of more open-weight models and proposing a framework for government-industry AI safety collaboration.
Sources: Meta AI Research, VentureBeat, CNBC
Jeff Dean Leaves Google After 27 Years to Co-Found Discovery Loop
In one of the most significant talent moves in AI history, Jeff Dean — Google's Chief Scientist and one of the most influential figures in modern computing — is leaving the company after 27 years to co-found Discovery Loop, a public benefit corporation focused on using AI to automate scientific and engineering research.
Joining Dean as co-founders are Sanjay Ghemawat (senior fellow), Quoc Le (founding member of Google Brain), and Oriol Vinyals (senior research scientist at Google DeepMind). Together, they represent a combined century of foundational AI research at Google.
Discovery Loop plans to use high-performance AI systems to initiate and iterate thousands of experiments simultaneously, partially automating the research process. Google itself will serve as a founding investor and cloud partner, suggesting the departure is amicable — but the loss of four key researchers in one move represents a significant reshuffling of AI talent.
The departure also triggered a broader leadership reshuffle at Google, with reports indicating that Demis Hassabis stepped down as DeepMind CEO as part of the reorganization.
Sources: TechCrunch, Bloomberg, CNBC
OpenAI's IPO Watch: S-1 Filed, September Timeline Wobbles
OpenAI's march toward becoming a public company took another step forward — and sideways. The company filed a confidential S-1 with the SEC earlier this summer, initially targeting a September 2026 listing at a valuation of up to $1 trillion, led by Goldman Sachs and Morgan Stanley.
However, CFO Sarah Friar told employees at an August all-hands meeting that OpenAI is "running its own race," and reports suggest the timeline may slip to 2027. The financials paint a complex picture: nearly $6 billion in Q1 revenue alone and a ~$25 billion annualized run rate, but also a staggering negative 122% non-GAAP operating margin — a loss of $1.22 for every dollar earned.
With more than 230 million weekly ChatGPT users and the Astra breakthrough generating headlines, OpenAI's IPO remains one of the most closely watched events in tech finance. Whether it lands in September or early 2027, it will test public market appetite for a company burning cash at an extraordinary rate while pushing the boundaries of AI capability.
Sources: CNBC, Evermx, Fortrove Partners