A developing picture of China’s AI, one story at a time.
Alibaba's AI coding assistant Qoder has launched a new Fast mode that cuts first-response time from 8 seconds to 3 seconds and reduces credit consumption by 27%, available across personal, enterprise, CN and international editions.
Chinese AI startup Shengshu Technology has released a preview of Vidu Q4, its next-generation video generation model supporting 4K/10-bit output, 16-second generation, and expanded multimodal reference inputs of up to 15 images and 3 audio tracks, positioning China's domestic video models closer to film-grade production quality.
Lenovo TianxiAI's self-developed code agent framework TianxiCode achieved a 71% problem-solving rate on the SWE-bench-Live Lite benchmark, ranking first globally with Verified certification. Running on DeepSeek-v4.1-Flash as its foundation model, the result highlights the growing competitiveness of Chinese-developed AI coding agents on international engineering benchmarks.
ByteDance updated its Doubao Work productivity platform on October 9 with an infinite canvas feature for multimodal creation, integration of the Seedream 5.0 Flash image generation model, and launch of Doubao 2.1 Lite optimized for office tasks. The update deepens multimodal AI integration in Chinese enterprise productivity tools.
Chinese EV maker XPeng has officially branded its robotaxi service XPENG YOYO and opened invitation-based public testing via a ride-hailing mini program in Guangzhou. The move marks a shift from internal validation to real-world user trials, using a vision-only second-generation VLA model powered by four in-house Turing AI chips delivering 3,000 TOPS. The company targets single-vehicle profitability in Guangzhou by the second half of 2027.
ComfyUI has demonstrated that MiniMax H3, a video generation model from Shanghai-based Chinese AI lab MiniMax, can produce live video at real-time speed on a single RTX 5090 consumer GPU through engineering optimizations including sparse attention, INT8 quantization and a fused FP4 MLP, reducing VRAM requirements from 80GB to under 30GB and making Chinese AI video generation accessible beyond expensive cloud infrastructure.
