Daily AI Updates: November 10, 2025

Main Content

• Performance optimization for the Kimi K2 Thinking model: Offers official API usage recommendations, including enabling streaming mode, setting the temperature to 1.0, and suggesting maximum tokens of 128k for inference and 256k for encoding, to enhance benchmark consistency and accuracy.
• Launch of the seventh-generation TPU chip: Delivers a 4x performance boost, supports AI model training and inference, and includes a major collaboration agreement with Anthropic.
• Version update for the data sharing feature: Postponed to the 2025.3.1 release, incorporating feedback to enhance security measures and streamline the opt-out process for non-commercial users, enabling better AI functionality integration.
• Expanded support for the Copilot agent: Introduces custom MCP server capabilities, improving the agent's intelligence and performance when handling additional context.
• Integration update for Copilot CLI: Rolls out new integrations and custom agent features, facilitating smarter building, collaboration, and deployment workflows.
• Integration optimization between Copilot and Playwright: Designed for validating UI changes, including app launches, interactive operations, and screenshot captures, to boost development efficiency.
• Power support for the Kimi K2 Thinking model: Leverages SGLang on Atlas Cloud to provide enhanced capabilities, improving model inference efficiency.