Kimi K3: China's 2.8-Trillion-Parameter Giant Tops Coding Benchmarks
The biggest story of the weekend belongs to Moonshot AI, the Beijing-based startup whose Kimi K3 model has sent shockwaves through the AI industry. At 2.8 trillion parameters, it is the largest open-weight model ever released — and it has claimed the top spot on major coding leaderboards, outperforming both Claude and GPT-series models on agent-programming scenarios.
Built on Moonshot's new Kimi Delta Attention (KDA) architecture — a hybrid linear attention mechanism designed for long-context efficiency — K3 ships with a 1-million-token context window, native multimodal input, and a reasoning mode that runs by default on every query. API pricing sits at roughly $3 per million input tokens and $15 per million output tokens, putting it in direct competition with frontier models from Anthropic and OpenAI.
Demand has been so overwhelming that Moonshot AI has temporarily stopped accepting new subscriptions, with available compute pushed to near capacity. The full open-source weights are scheduled for release on July 27, alongside a technical report — a date the entire industry is watching closely.
Source: VentureBeat, Fortune
Google Delays Gemini 3.5 Pro for a Third Time
Google is having a brutal week. According to Bloomberg, the company has delayed the public launch of Gemini 3.5 Pro — its most powerful flagship AI model — for the third time, after internal testing revealed the system fell short on coding performance and complex, long-horizon reasoning tasks.
Originally promised at Google I/O alongside the already-launched Gemini 3.5 Flash, the Pro variant was supposed to arrive in June. Google reportedly updated its training data in late June to improve coding capabilities, but the results were disappointing. The model remains in limited enterprise preview on Vertex AI, and the company is reportedly exploring a stopgap Gemini 3.6 Flash release while it works on fixing Pro.
Alphabet shares fell approximately 4% on the news, adding financial pressure to the technical setback. The delay comes at the worst possible moment: Kimi K3 is surging, Anthropic's Claude Sonnet 5 is gaining traction, and OpenAI's GPT-5.6 family is already in the market.
Source: 9to5Google, Business Today
EU Orders Google to Open Android to Rival AI Assistants
The European Commission has adopted two binding decisions under the Digital Markets Act that could reshape the AI assistant landscape on the world's most popular mobile platform. Google must now give competing AI assistants — from Anthropic, OpenAI, and others — the same system-level access on Android that it currently reserves for its own Gemini assistant.
The first decision requires Google to enable users to choose a preferred AI assistant, activate it by voice, and let it perform tasks inside apps — booking rides, suggesting replies in messaging apps, answering questions about recently visited places. Full Android interoperability compliance is due by July 2027.
The second decision mandates that Google share anonymized search data with competing search engines and AI chatbots with search features, beginning January 2027. This targets the data advantage Google has built over decades — the kind of scale that no competitor can replicate on their own.
The ruling has broader implications: Apple is facing similar pressure over Siri's integration on iOS, and CNN reports both companies are pushing back against the scope of the requirements.
Source: European Commission, Unite.AI
SAP Completes €1 Billion Prior Labs Acquisition, Bets on European Frontier AI
SAP has officially completed its acquisition of Prior Labs, the German AI startup behind tabular foundation models, committing more than €1 billion over four years to scale it into a globally leading frontier AI research lab based in Europe.
The deal is remarkable for its speed: Prior Labs was founded just 18 months ago. Its flagship TabPFN model series, published in Nature, set the state-of-the-art on tabular data benchmarks — the kind of structured data (spreadsheets, databases, business records) that underpins enterprise operations worldwide.
While the rest of the industry chases larger language models and multimodal capabilities, SAP is making a deliberate bet that purpose-built models for structured data represent a distinct and underserved frontier. Prior Labs will continue to operate independently, and the investment positions Europe with its first major frontier AI lab since Mistral.
Source: SAP News Center, Tech.eu
South Korea Commits $880 Billion to Semiconductors and AI Infrastructure
South Korean President Lee Jae-myung has unveiled a ₩1,350 trillion (approximately $880 billion) 10-year public-private investment plan covering semiconductors, AI data centers, and robotics — the largest national AI infrastructure commitment announced by any country to date.
The plan centers on Samsung Group and SK Group, which will build four new chip fabrication plants in the country's southwest. SK Group, GS Group, and Naver will invest 550 trillion won to build AI data centers, targeting 8.4 gigawatts of capacity by 2029 and another 10 gigawatts by 2035.
The sheer scale underscores a global consensus: the AI race will increasingly be won in hardware and infrastructure, not just model architecture. South Korea is betting that its existing dominance in high-bandwidth memory (HBM) chips — essential for AI training and inference — gives it a structural advantage that software-first competitors cannot easily replicate.
Source: TechSpot, Al Jazeera
What to Watch This Week
Several important dates are ahead: DeepSeek V4 stable release is expected on July 24, followed by Kimi K3's full open-weight release on July 27. The newly formed WAICO (World AI Conference Organization) is expected to announce its charter and leadership. And reports suggest Anthropic's IPO process continues to advance, though no timeline has been confirmed.