Particle.news
Download on the App Store

China’s Tech Giants Roll Out AI Agents as OpenAI Reports Near-Expert Model Performance

Fresh benchmarks alongside product tests signal AI moving into commerce and daily life.

Overview

  • Alibaba’s 1688 unveiled the cross-border B2B agent “遨虾” in internal testing with a November 2025 launch target, promising visual and semantic matching of overseas products to Chinese suppliers plus early compliance prompts.
  • The 1688 AI app added “AI找厂” for factory matching and “AI参谋” for business analysis, framing an “AI industrial large model” built on 26 years of platform data to shorten sourcing cycles for small buyers.
  • JD showed its internal “京犀” app as an AI shopping and life entry that handles natural-language requests across retail, on-demand delivery, and travel using its Oxygen architecture with JoyAgent multi-agent orchestration.
  • Meituan’s “小美” entered public beta powered by its LongCat-Flash-Chat MoE model, aiming to complete local-life tasks from preferences and context with high token throughput and lower output cost.
  • OpenAI introduced the GDPval benchmark for report-style tasks across 44 professions, with GPT-5-high rated at or above experts in 40.6% of cases and Anthropic’s Claude Opus 4.1 at 49%, while noting the test’s limited scope and Sam Altman’s forecast of AGI before 2030 with AI eventually handling 30–40% of work.