VEO 3.1 – Google’s Answer to SORA 2

Google released VEO 3.1 for AI video generation with enhanced control and integrated audio, while Anthropic launched the faster and cheaper Claude Haiku 4.5, Kong introduced the MCP-native AI agent toolkit Volcano, and Andrei Karpathy provided NanoChat, a trainable ChatGPT-style model for about $100.

• Google's VEO 3.1 offers advanced AI video generation with integrated audio and enhanced creative control.

• Anthropic's Claude Haiku 4.5 delivers faster, cheaper, and safer AI performance for real-time applications.

• Kong's Volcano provides an open-source, MCP-native framework for efficient AI agent development.

Google unveiled VEO 3.1, its advanced AI video model, offering significant improvements in cinematic quality, physics, and full audio generation, now available via Flow and Gemini API with new editing tools. Concurrently, Anthropic released Claude Haiku 4.5, a highly efficient and safer model that is twice as fast and a third of the cost of its predecessors, ideal for real-time AI applications. Additionally, Kong launched Volcano, an open-source toolkit for building AI agents that natively supports the Model Context Protocol (MCP), and Andrei Karpathy made NanoChat available, a lightweight, customizable ChatGPT-style model that can be trained from scratch for a minimal cost.

Google's VEO 3.1

• 00:00:38 Google has launched VEO 3.1, an advanced AI video model enhancing cinematic quality, motion, and natural physics. It now integrates full audio generation, including dialogue and ambient sounds, available through Google's AI filmmaking platform Flow and the Gemini API. The model supports 720p or 1080p output at 24 frames per second, with an 'extend' feature allowing videos up to two and a half minutes with continuity.

Anthropic's Claude Haiku 4.5

• 00:04:56 Anthropic introduced Claude Haiku 4.5, a highly efficient and safer AI model designed for speed and cost-effectiveness. This compact model operates at double the speed and one-third the cost of Claude Sonnet 4.5, excelling in tasks requiring direct computer use like real-time assistants and coding co-pilots. Haiku 4.5 also achieved an AI safety level 2 certification, demonstrating lower rates of misaligned behavior than previous frontier models.

Kong's Volcano AI Agent Toolkit

• 00:07:37 Kong unveiled Volcano, an open-source toolkit for building smart AI agents that leverages the Model Context Protocol (MCP) natively, allowing seamless integration with various AI models and tools. A live demonstration showcased its ability to process over 900 coffee orders with just 15 lines of code, highlighting its efficiency and multi-agent orchestration capabilities. Volcano is fully open source and is intended to simplify AI agent development by providing a cleaner, faster framework.

Karpathy's NanoChat

• 00:10:05 Andrei Karpathy released NanoChat, a mini ChatGPT-style model that can be trained from scratch in approximately four hours for under $100, providing an accessible open-source learning tool. This lightweight setup, comprising about 8,000 lines of code, manages training, fine-tuning, testing, and chatbot operation through a web interface. NanoChat trains in three steps: general language patterns, conversations/math using Python, and fine-tuning on high-quality chat data, with an optional fourth step for learning from mistakes, offering transparent insights into LLM functionality.