Chinese startup Moonshot AI has unveiled Kimi K3, a massive 2.8-trillion parameter model featuring native multimodal capabilities, a 1 million token context window, and a novel Kimi Delta Attention mechanism. The model's launch, which claims parity with top US models in coding and agentic tasks, triggered a massive global sell-off in AI and semiconductor stocks, wiping out $950 billion in US equity value and allowing Apple to unseat Nvidia as the world's most valuable company.
The newly released Kimi K3 AI model has achieved top rankings across multiple benchmarks, including SpreadsheetBench 2, creative writing, and the Frontend Code Arena, where it helped China surpass the US in frontend code generation for the first time. Analysis shows Kimi K3 matches the software engineering performance of Claude Fable 5 at approximately 35% of the cost. This rapid advancement has intensified debates around US regulatory policies and the global balance of AI leadership.
Meta is reportedly in early discussions to lease up to $10 billion in AI computing power to Anthropic over two years, potentially launching Meta's first dedicated cloud computing business. To bolster its expanding infrastructure efforts, Meta has also hired former Amazon Web Services senior executive Dave Brown to join its data center and compute team.
OpenAI has officially released its new GPT-5.6 model family, which includes three distinct sizes named Luna, Terra, and Sol. These models feature multiple reasoning-effort settings, marking a significant advancement in LLM-based reasoning and establishing it as a standard component of modern AI development.
Anthropic has reversed its decision to remove its advanced Claude Fable 5 model, making it a permanent feature for Max and Team Premium subscribers starting July 20 due to competitive pressure. To manage high demand, access will be capped at 50% of full usage limits. Additionally, the company is issuing refunds and $100 credits to users affected by a recent software bug that temporarily blocked access to the model.
The Trump administration has launched the 'Gold Eagle' cybersecurity clearinghouse, which requires government approval for early access to frontier AI models from developers like OpenAI and Anthropic. Additionally, the administration is planning to establish an independent, FINRA-style regulator reporting to the SEC to standardize AI model safety vetting. Together, these initiatives shift significant oversight and distribution control of cutting-edge AI from tech companies to federal authorities.
Chinese President Xi Jinping announced the formation of the World Artificial Intelligence Cooperation Organization (WAICO) with 29 member states, committing to provide 5,000 AI training opportunities for developing countries. This move positions China to lead global AI governance, advocating for open-source models and human oversight while challenging Western-centric frameworks.
Nvidia will supply 27,500 Rubin GPUs and 13,750 Vera CPUs to Noetra, a Japanese government-backed consortium, for a 140 MW AI data center. This facility will serve as the central compute hub for a national physical AI project aiming to deploy 10 million AI-equipped robots in Japan by 2040, backed by $2.4 billion in public funding.
NVIDIA and Hugging Face have announced an integration between NVIDIA NeMo Automodel and the Hugging Face Diffusers library to enable scalable, production-grade training for large diffusion models. This collaboration allows developers to fine-tune models directly from the Hugging Face Hub using advanced sharding and parallelisms, with plans for a Pythonic API to further simplify workflows. Separately, NVIDIA's Nemotron 3 Embed model achieved the top ranking on the Retrieval-Augmented Generation Evaluation Benchmark.
Google Vids has introduced major updates including Gemini Omni integration for natural language video generation and editing, alongside a new feature allowing users to create personalized digital avatars for narration using a selfie and voice recording. To ensure transparency and responsible AI use, all AI-generated video clips in the platform will now feature invisible SynthID digital watermarks.
Amazon has introduced Amazon Quick, an AI-powered assistant designed to boost sales productivity by automating administrative tasks such as CRM updates, email drafting, and lead scoring. The tool features no-code application building, relationship mapping via a Knowledge Graph, and automated call transcript scoring against methodologies like MEDDPICC. By identifying deal risks and recommending next actions, Amazon Quick aims to streamline sales cycles and reduce manual research for sales representatives.
Amazon Web Services has announced the general availability of Amazon Bedrock's Managed Knowledge Base, a fully managed solution designed to simplify the creation of generative AI applications through agentic retrieval. The service features an Advanced Agentic Retrieval API for complex query decomposition, native enterprise connectors with real-time access control, and fully managed multimodal data parsing. Notably, OpenAI is adopting these new capabilities to enhance and scale its Retrieval Augmented Generation (RAG) systems for millions of users.
xAI has made its Grok 4.3 large language model generally available on Amazon Bedrock, offering advanced reasoning capabilities, a 1 million token context window, and strong tool use for enterprise and agentic workloads. This partnership integrates Grok 4.3 into Amazon's cloud AI service, providing developers with a powerful new model for building AI applications.
Anthropic has reportedly limited the rollout of its new AI model, Claude Fable 5, with venture capitalist Chamath Palihapitiya citing compute shortages as the reason. This highlights the increasing challenge of securing sufficient computational resources for deploying advanced AI models across the industry.
The departure of a Moonshot AI founder has ignited discussions regarding US immigration policy for AI talent, specifically a proposal by David Pfau for automatic O-1 visas for Machine Learning PhDs. This highlights challenges in retaining top AI talent in the US.