Google Gemini 3.5 Pro Arrives After Full Architectural Rebuild
After scrapping its original base model and restarting pretraining from scratch, Google DeepMind has launched Gemini 3.5 Pro — the company's most ambitious model release in over a year. The new model ships with a 2-million-token context window, doubling the capacity of any other frontier model currently available, and introduces a new Deep Think extended reasoning mode gated behind the $250/month Ultra subscription.
Google engineers reportedly scrapped the original Gemini 3.5 Pro after discovering structural failures in recursive tool-calling and SVG generation, opting for a complete architectural rebuild rather than patching the existing system. The result is a model targeting enterprise and research users where raw capability matters most, with enhanced reasoning, faster inference, and support for autonomous workflows that can manage coding tasks with minimal human intervention.
The timing is no accident. By launching on the same day as the World AI Conference in Shanghai, Google is making a pointed statement about its position in the intensifying US-China AI competition. TechTimes | Startup Fortune
Xi Jinping Delivers First-Ever Keynote at World AI Conference as 29 Nations Form New AI Body
Chinese President Xi Jinping made history on Thursday by personally delivering the keynote address at Shanghai's World AI Conference — his first appearance at the event since it began in 2018. Traditionally, China's premier has attended instead, so Xi's presence signals a dramatic elevation of AI's priority within China's political hierarchy.
"AI development should not be a solo performance by a single country, but a symphony of international cooperation," Xi told over 1,400 attendees including executives, investors, academics, and representatives from 12 government ministries. He called for safety risks to be contained while emphasizing open collaboration — a message pointedly directed at Washington's export-control approach.
The conference's most consequential outcome: 29 countries — including Russia, Brazil, Pakistan, Indonesia, and Serbia — signed an agreement to establish the World Artificial Intelligence Cooperation Organization (WAICO), an independent intergovernmental body headquartered in Shanghai. The move positions China as the architect of an alternative AI governance framework, offering developing nations "open weights, cheaper models, and a seat at a table that Washington has not offered them." The Next Web | CGTN
AI Safety Index 2026: No Frontier Lab Scores Above C+
The Future of Life Institute released its 2026 AI Safety Index this week, and the results paint a sobering picture of the industry's safety posture. No model developer scored higher than a C+ — and several earned outright failing grades.
Anthropic again leads the pack at C+, maintaining its position through relatively strong transparency, safety frameworks, and governance practices. OpenAI dropped from last year's C+ to a C, while Google DeepMind held steady at C. Meta showed improvement, climbing from 6th to 4th place with a D+. At the bottom: xAI plunged from 4th to 7th with an F, DeepSeek fell to F, and Mistral scored the lowest overall at F.
The most alarming finding: every company scored D or below on existential safety, the weakest domain across the industry. The report also flagged a troubling trend where labs that once banned military applications — including Anthropic, OpenAI, Google DeepMind, and Meta — have reversed course and are now actively seeking defense partnerships. Future of Life Institute
Moonshot AI Drops Kimi K3: First Open Model in the 3-Trillion-Parameter Class
Chinese AI lab Moonshot AI officially released Kimi K3, making it the first open model to cross the 3-trillion-parameter threshold. The 2.8T-parameter model features Kimi Delta Attention, a 1-million-token native vision context window, and is available immediately in Kimi App, Kimi Work, Kimi Code, and via API. Full model weights are expected to drop on July 27.
The release underscores China's accelerating push into open-weight frontier models — a strategy that directly supports Beijing's geopolitical positioning as an alternative to US-dominated closed-source AI. Coming on the same day as the World AI Conference, it reinforces the narrative that Chinese labs can compete at the frontier while maintaining a more open approach.
200+ Experts Including 16 Nobel Laureates Demand Urgent Action on AI's Economic Impact
Over 200 economists and AI researchers — including 16 Nobel laureates — released a statement titled "We Must Act Now" warning that AI's economic transformation could be "larger than the Industrial Revolution, but unfolding over a vastly shorter time frame."
The statement, organized by Stanford's Erik Brynjolfsson and colleagues, was signed by heavyweights including OpenAI finance chief Sarah Friar, Google DeepMind chief scientist Jeff Dean, and Anthropic co-founder Jack Clark. It warns of large-scale job displacement alongside potential major gains in living standards, calling on policymakers to begin building institutions now rather than after the disruption has already hit.
Nobel laureates Daron Acemoglu, Simon Johnson, and Michael Spence are among the signatories, lending significant institutional weight to what has often been dismissed as speculative concern about AI's labor market impact. Stanford Digital Economy Lab | The Washington Post