AI news, 21 July: Chinese models challenge US lead as Shanghai summit closes

Chinese labs unveil frontier models and jolt the AI race. Over the weekend and into Monday, Chinese companies drew intense attention with major model announcements. Moonshot AI’s Kimi K3, a 2.8-trillion-parameter system billed as the largest open-weight model yet, has topped at least one coding leaderboard and is slated for downloadable weights on 27 July, according to coverage by BBC and Artificial Intelligence News. Alibaba separately previewed Qwen3.8 (about 2.4 trillion parameters), claiming it ranks second only to Anthropic’s Claude Fable 5; The Wall Street Journal and South China Morning Post reported the claim and a subsequent rise in Alibaba’s Hong Kong-listed shares. Independent verification of the full benchmark picture is still limited, and both labs still place their systems behind or level with top U.S. models on some measures, but the releases underscore how quickly Chinese teams are closing the capability gap despite chip export limits.

World AI Conference ends with a new cooperation body. Shanghai’s World Artificial Intelligence Conference and High-Level Meeting on Global AI Governance wrapped on 20 July. Chinese President Xi Jinping had delivered a rare keynote urging open-source AI and broader access for developing countries. On the eve of the event, 29 countries signed on to create the World Artificial Intelligence Cooperation Organization (WAICO), framed as a venue for more inclusive global AI rules. Western governments and labs will watch whether the group becomes a genuine multilateral forum or mainly a China-led alternative to existing initiatives.

U.S. public-health agencies to pilot OpenAI and Anthropic tools. The Coalition for Health AI announced the PULSE programme on 20 July: ten state, local, tribal or territorial jurisdictions will test enterprise generative-AI tools from OpenAI and Anthropic, with Accenture supporting onboarding. Planned use cases include disease surveillance, community-health mapping, multilingual public communication and automated clinical-data queries. Playbooks for wider use are expected in 2027. Organisers stress governance and privacy, though detailed evaluation rules and data-handling practices have not yet been published.

“Vibecoded” apps flood Apple’s App Store. The New York Times reported on 20 July that AI coding tools have turbocharged app creation: new App Store submissions roughly doubled in the first half of 2026 versus a year earlier, per Sensor Tower estimates, while overall downloads rose only modestly. Many makers build by describing ideas to models rather than writing code by hand. Apple says most apps are still reviewed within 48 hours and welcomes the new generation of developers; critics worry about clutter, quality and longer review queues.

Politicians push to shape what chatbots say about them. Also this week, The New York Times detailed how campaigns and office-holders are pressing AI companies and deploying their own tools so that chatbot answers about candidates are more favourable or complete. The episode highlights a growing public fight over whether commercial AI systems should act as neutral encyclopaedias, or whether political actors can and should influence the narratives voters see when they ask everyday questions.