How to access Chinese LLM chatbots across the world (www.technologyreview.com)

Chinese LLM chatbots are far more accessible to international users than they might seem: while a Chinese phone number still unlocks the widest set of services (some models from Huawei, 360, and the Chinese Academy of Sciences remain gated), many notable systems support English and can be tested without such a number. Workarounds include direct sign-ups for platforms like ByteDance’s Doubao, Zhipu’s ChatGLM, and startup DeepSeek (which offers very low token pricing — reported at $0.14 per million tokens vs. GPT-4-Turbo’s ~$10 per million), plus global hubs like Hugging Face and Alibaba’s ModelScope that don’t require domestic verification. Hugging Face hosts demos and code for models such as Alibaba’s Qwen (which recently topped its Open LLM Leaderboard) and Tencent’s image model Hunyuan-DiT, letting developers and nontechnical users try text and multimodal features instantly. ModelScope’s LLM Arena (from Shanghai’s Open Compass) integrates about 11 Chinese models — including Baichuan, Baidu Ernie Bot, iFlytek Spark, Minimax, Moonshot, and InternLM — enabling side-by-side prompt comparisons (though some multimodal functions like image upload/generation aren’t available via these demos). The takeaway: global researchers and practitioners can now more easily benchmark, probe, and compare Chinese LLMs, accelerating cross-ecosystem evaluation despite remaining geo-restrictions and feature limitations.