Ad
Short

Alibaba has developed a new AI chip, which is currently in testing, designed for a broad range of inference tasks, such as powering the responses from a smartphone voice assistant. The chip is manufactured by a Chinese company and is more versatile than Alibaba's older chips. It is designed for inference, not for training AI models—an area where China's biggest weakness lies compared to the US.

As reported by the Wall Street Journal, Alibaba's new chip is compatible with the Nvidia software platform, meaning engineers can repurpose programs written for Nvidia hardware. The chip helps to fill the void created after Nvidia ran into regulatory barriers restricting sales of its products in China. Alibaba was long one of Nvidia's biggest customers before these restrictions were put in place.

Short

X.AI has introduced Grok Code Fast 1, a new AI model designed specifically for agent-based programming. The company says it uses a "new architecture," was trained on real-world programming data, and is built to be fast and cost-effective. Grok Code Fast 1 supports a wide range of programming languages and is intended to handle tasks like bug fixes and project setup on its own. Early integration partners include GitHub Copilot, Cursor, and others, where the model is available for free testing for a limited time. Pricing is set at $0.20 per million input tokens and $1.50 per million output tokens. X.AI hasn't released benchmark comparisons, sharing only a single SWE-Bench score of 70.8 percent. According to initial user reports, the model is fast, but struggles with more complex tasks and makes frequent mistakes. Grok Code Fast 1 seems positioned as an alternative to smaller models like GPT-5-nano.

Short

Anthropic is rolling out new data privacy controls for Claude. Users on the Free, Pro, and Max plans must now actively opt out if they don't want their conversations used to train AI models.

The new setting only applies to new or ongoing chats and can be changed at any time. If you allow data use, Anthropic will keep your chat data for up to five years to help improve its models and security systems. If you opt out, your conversations are stored for just 30 days. These changes don't affect Claude for Work, Education, Government, or API access through partners like Amazon Bedrock.

Users have until September 28, 2025, to make their choice. After that, you'll have to select a data sharing preference to keep using Claude.

Short

Google's Flow tool now gives users a choice each month: five free Veo 3 Fast AI videos or one standard Veo 3 video. Google never spelled out the quality gap between the two models, but the new credit system offers a clue: one standard video costs as much as five Fast videos. Every user gets 100 free credits per month, enough for either option. Flow supports scene editing with AI ensuring consistency. For developers, API pricing starts at $0.040 per second for Fast and $0.0075 per second for standard videos.

Google News