AI news, 16 July: Moonshot unveils massive open model closing gap with US leaders

Moonshot launches Kimi K3 as world's largest open-weight model. Chinese startup Moonshot unveiled Kimi K3, a 2.8-trillion-parameter open-weight AI model it describes as approaching the performance of leading US systems such as Anthropic’s Fable. According to Reuters and company statements, the model features a 1-million-token context window for advanced reasoning, coding and long tasks. Third-party benchmarks ranked it highly, including first in one frontend-coding arena and second overall on another index, sometimes ahead of or competitive with models like GPT-5.x and Claude Opus variants. Open-weight status lets users download and customise it. Analysts note Chinese models are gaining ground and running more cheaply, though hosting such a large system remains costly for most. The release underscores accelerating progress outside US closed labs.

29 countries create World AI Cooperation Organization. On the eve of Shanghai’s World AI Conference, 29 countries including Russia, Brazil, several African and Asian nations signed an agreement establishing the China-proposed World AI Cooperation Organization (headquartered in Shanghai), Reuters reported. China frames it as promoting international cooperation and governance. The signing involves founding members from the Global South and partners such as Belarus, Cuba and Venezuela; Western powers including the US are not listed among them. President Xi Jinping is expected to outline China’s vision for AI rules at the conference opening. This marks a concrete step after earlier proposals, positioning China to influence global AI norms amid US-China tech rivalry.

EU orders Google to open Android and search data to rivals. European regulators detailed requirements forcing Google to grant competing AI systems (including OpenAI’s) equal access to key Android features so they can better challenge Gemini, and to share anonymised search data with rivals starting in stages from 2027, according to Reuters and other reports. These DMA-related measures aim to curb Google’s advantages. Google has expressed concerns about privacy and security impacts. The rules take effect gradually, giving competitors practical tools to integrate more deeply into Android.

Thinking Machines Lab releases first open model Inkling. Former OpenAI CTO Mira Murati’s startup Thinking Machines Lab shipped Inkling, an open-weights multimodal Mixture-of-Experts model with roughly 975 billion total parameters (about 41 billion active), a 1-million-token context, and native support for text, images, audio and video, according to the company, TechCrunch and WSJ. Trained from scratch on a large mix of data, it is designed for customisation, reasoning and collaboration rather than pure scale. The release gives developers a strong Western open alternative and comes after the company spent time building infrastructure out of the public eye.

Neko Health raises $700 million for AI body scans. The Swedish preventive-health company co-founded by Spotify’s Daniel Ek secured $700 million in Series C funding (led by Lightspeed and O.G. Venture Partners, with high-profile individual backers) to expand its AI-powered full-body scanning clinics into the United States, with a New York site planned, Bloomberg and company announcements said. The technology aims to spot health risks early through comprehensive non-invasive scans and AI analysis. The large round reflects investor confidence in AI applications that blend hardware, sensors and models for personalised preventive care.