Apple Launches M6 Mac Mini and M5 Ultra Mac Studio; OpenAI Chip and Driving Law in Focus

The Mac mini is available in M6 and M5 Pro versions, priced at 6,999 yuan and 12,999 yuan, while the Mac Studio comes with M5 Max or M5 Ultra, starting at 19,999 and 46,999 yuan. Pre-orders in China begin August 27, with sales starting September 22. The M6 integrates a 12-core CPU, 12-core GPU and dual 16-core neural engine, with memory bandwidth up to 170GB/s. The M5 Ultra uses a quad-die architecture and offers up to a 36-core CPU, 80-core GPU, 512GB unified memory and 1.2TB/s memory bandwidth, delivering up to 4.3 times the AI peak performance of the M3 Ultra.

A draft amendment to China's Road Traffic Safety Law was submitted for initial review on August 25. For the first time, it includes a chapter on autonomous vehicles, proposing that when a vehicle is in autonomous driving mode, the manufacturer or importer will handle traffic violations. If autonomous functions are disabled or the vehicle only has assisted driving, ordinary rules apply. The draft also addresses road occupation, e-bike speed limits and mandatory helmet use.

OpenAI released initial benchmark results for its first in-house reasoning chip, Jalapeño, showing it outperforming Nvidia's GB200 and GB300 in throughput, energy efficiency and latency. On GPT-OSS 120B, it reached 1,459 tokens per second, 2.7 times the GB200's rate. On DeepSeek R1 670B, end-to-end latency was 1.65 seconds versus 5.99 seconds for the GB300, and per-kilowatt throughput was 104.3 times higher. Overall AI work per watt improved 1.5 to 1.9 times. The chip consumes 700W, half the GB300's power. OpenAI said it took nine months from design to tape-out, with AI assistance cutting SIMD unit area by 8% and matrix engine area by 10%. The chip is the first in a roadmap co-developed with Broadcom, and OpenAI plans to deploy it in its own infrastructure by year-end.

Volkswagen is pursuing what CEO Oliver Blume described as the largest transformation in the group's history, potentially cutting about 50,000 additional jobs and spinning off some business units, according to Reuters and Jiemian News. The proposals require further negotiation with the supervisory board and unions, which have publicly questioned them. Volkswagen aims to reduce costs, simplify models and align capacity closer to actual demand of about 9 million vehicles per year.

ByteDance launched Doubao Work, an AI agent that can operate a virtual desktop and read context from Feishu, its enterprise communication platform.