Moonshot AI releases Kimi K3, a 2.8 trillion parameter open MoE model

Moonshot AI has released Kimi K3, a 2.8 trillion parameter open MoE model with native vision and a 1 million token context window. This model targets long-horizon coding, knowledge work, and reasoning, outperforming other tested models.

Kimi K3 ranks second in internal evaluations, behind only Claude Fable 5 and GPT 5.6 Sol. This achievement could put pressure on OpenAI and Anthropic, as Kimi K3 offers native visual understanding and a 1M-token context window.

Nvidia is partnering with Toyota's Woven City in Japan to accelerate the adoption of artificial intelligence. The partnership will provide foundational technologies for Woven City, a smart city project.

The US Navy has released a new strategy, the Bits2Effects Cycle, to deploy data and artificial intelligence. The plan aims to transform raw information into meaningful battlefield outcomes, enabling the Navy to out-learn and out-fight any adversary.

Researchers have identified a new class of attack called agent data injection (ADI), which can make AI agents misclick or run attacker commands. ADI was tested on six AI models, including OpenAI's GPT-5.2 and GPT-5-mini.

The Thinking Machines Lab has released Inkling, a 975-billion-parameter open source model that understands video and audio. Inkling can be used in various applications, including natural language processing and computer vision.

Demis Hassabis and Wendy Hall emphasized the importance of education and global governance in the evolving landscape of artificial intelligence. They discussed the critical need for a comprehensive national AI strategy.

Key Takeaways

['Moonshot AI releases Kimi K3, a 2.8 trillion parameter open MoE model.', 'Kimi K3 ranks second in internal evaluations, behind Claude Fable 5 and GPT 5.6 Sol.', "Nvidia partners with Toyota's Woven City to accelerate AI adoption.", 'US Navy releases Bits2Effects Cycle strategy to deploy data and AI.', 'Researchers discover agent data injection (ADI) attack, tested on OpenAI models.', 'Thinking Machines Lab releases Inkling, a 975-billion-parameter open source model.', 'Demis Hassabis and Wendy Hall emphasize education and global governance in AI.', 'Burkina Faso trains lawmakers on AI for digital governance.', 'US targets far-left terror groups with visa bans.', 'Patter SDK guide provides tutorial on building voice-agent workflow.']

Moonshot AI Unveils 2.8 Trillion Parameter Model

Moonshot AI has released Kimi K3, a 2.8 trillion parameter open MoE model with native vision and a 1 million token context window. It uses Kimi Delta Attention and Attention Residuals to improve information flow. The model targets long-horizon coding, knowledge work, and reasoning. Kimi K3 is the first open model to reach 2.8 trillion parameters and outperforms other tested models.

Kimi K3 Ranks Second in Internal Evaluations

Moonshot AI's Kimi K3 model ranks second in internal evaluations, behind only Claude Fable 5 and GPT 5.6 Sol. Kimi K3 has 2.8 trillion parameters and uses Kimi Delta Attention and Attention Residuals. The model offers native visual understanding and a 1M-token context window. It could put pressure on OpenAI and Anthropic.

New Agent Data Injection Attack Discovered

Researchers have identified a new class of attack called agent data injection (ADI), which can make AI agents misclick or run attacker commands. ADI works by corrupting the small facts an agent quietly trusts. The attack was tested on six AI models, including OpenAI's GPT-5.2 and GPT-5-mini.

Nvidia Partners with Toyota's Woven City

Nvidia is working with Toyota Motor's Woven City, a smart city in Japan, to speed up the spread of artificial intelligence. The partnership will provide foundational technologies for Woven City.

US Navy Unveils AI Strategy

The US Navy has released a new strategy to deploy data and artificial intelligence. The plan, called the Bits2Effects Cycle, aims to transform raw information into meaningful battlefield outcomes. The Navy wants to out-learn and out-fight any adversary.

Thinking Machines Lab Releases Inkling Model

The Thinking Machines Lab has released its first model, Inkling, a 975-billion-parameter open source model that understands video and audio. The model can be used in various applications, including natural language processing and computer vision.

US Targets Far-Left Terror Groups with Visa Bans

The US Department of State has announced a new visa restriction policy targeting individuals associated with far-left terrorist groups. The policy aims to counter the use of organized networks and violence as a political tool.

Demis Hassabis & Wendy Hall on AI's Future Skills Gap

Demis Hassabis and Wendy Hall discussed the evolving landscape of artificial intelligence and the critical need for a comprehensive national AI strategy. They emphasized the importance of education and global governance.

AI Under Pressure

The use of AI is transforming our world, but it also raises concerns about scams, security, and sustainability. As AI becomes more prevalent, so do the risks associated with it.

Patter SDK Guide

The Patter SDK guide provides a tutorial on building a voice-agent workflow that simulates how an AI phone assistant behaves during real conversations.

Burkina Faso Trains Lawmakers on AI

Burkina Faso has trained lawmakers on AI to support digital governance. The training aims to help lawmakers understand AI and develop legislation that promotes responsible AI governance.