Evum AI logo
← Back to newsletterSeptember 14, 2026

GPT-6 Astra Debuts & AI Cracks Millennium Math Problems

Frontier AI labs launched a new generation of enterprise-ready agentic models this week, alongside astonishing breakthroughs in mathematics where models solved and verified historically challenging theorems.

🤖 OpenAI Launches GPT-6 Astra with Desktop Enterprise Integration

OpenAI has released GPT-6 Astra, its most capable model to date, featuring built-in browser-use capabilities. The launch includes enterprise plugins for tools like Oracle Analytics and Power BI, allowing the model to interact directly with standard business software.

openai.com →

🧮 Anthropic Releases Claude Fable 5.1 and Claude Mythos 5.1 Models

Anthropic has launched its next-generation Claude 5.1 models, Fable and Mythos, designed to push boundaries in complex reasoning. The system card highlights evaluations on ArXivMath, a dynamic benchmark extracted from active, unresolved research papers to measure true frontier math capabilities.

www-cdn.anthropic.com →

🔬 Internal Claude Model Discovers Landmark Proof on Riemann Zeta Function Zeros

An internal research version of Anthropic's Claude has formulated a breakthrough mathematical proof regarding the non-trivial zeros of the Riemann zeta function. The AI-generated proof, which shows more than 67.25% of the zeros are simple and on the critical line, was subsequently verified by independent mathematicians.

arxiv.org →

📐 OpenAI and Anthropic Researchers Clash on Navier-Stokes Millennium Problem

OpenAI and Anthropic teams concurrently pursued computational resolutions to foundational fluid dynamics problems, with OpenAI targeting the Navier-Stokes equations and Anthropic focusing on the forced Euler problem. The concurrent discoveries underscore the escalating race between frontier labs using automated proof assistants and LLMs.

openai.com →

🧬 Google DeepMind Releases AlphaGenome Mapping 9 Billion Gene Mutations

Google DeepMind has launched AlphaGenome, an open-access atlas predicting the functional effects of all 9 billion possible human gene mutations. The platform has already seen broad adoption from thousands of biologists, providing a highly scalable computational resource for disease diagnosis and drug discovery.

nature.com →

🏢 Heavy API Price Cuts Challenge Anthropic’s IPO Valuation Narrative

Intense price cuts across the LLM industry are raising concerns among investors regarding Anthropic’s revenue projections as it prepares for an initial public offering. The downward pricing pressure threatens to commoditize basic model intelligence, forcing labs to find high-margin enterprise value elsewhere.

theinformation.com →

🏭 China’s CXMT Starts Small-Scale Production of Advanced High-Bandwidth Memory

ChangXin Memory Technologies (CXMT), China's premier memory chipmaker, has successfully begun producing high-bandwidth memory (HBM) in limited quantities. This breakthrough marks a crucial step in China's drive for semiconductor self-sufficiency amidst mounting U.S. export controls on AI-critical hardware.

theinformation.com →

🌍 Military Drone Data from Ukraine Spurs Unregulated Defense AI Market

Drone footage and telemetry collected from active combat in Ukraine are being packaged to train next-generation autonomous military hardware. This rapid accumulation of real-world battle data is occurring within a global legal vacuum, sparking debate over consent, data governance, and weaponization.

technologyreview.com →

🧠 Deep Learning Models Enable Real-Time Emotion Decoding from Brain Activity

Researchers have developed personalized deep learning models that can decode continuous human emotional states directly from intracranial brain activity. By processing signals from both gray and white matter, the low-latency system successfully generalized across complex visual tasks, opening new avenues for neural interfaces.

nature.com →

🚀 OpenAI Pivots Strategy Toward Recursive Self-Improvement and Automated Research

OpenAI has detailed its focus on recursive self-improvement (RSI), arguing that automated AI agents are the most scalable path to advance the scientific frontier of AI. By designing systems that can write, evaluate, and align their own successor models, the company aims to dramatically accelerate development.

openai.com →