The capabilities, trade-offs and people behind China’s AI models.
In a November 8 original report on a public dialogue, Yicai recounts Tencent scientist Yao Shunyu’s view that user context and trust can matter more than model leadership for lasting AI product advantages.
A November 7 QbitAI hands-on report compares AFK AI's Comma with DeepSeek's Super-agent architecture proposal, describing parallel task execution and Linear integration while separating product history from the new reporting.
Tencent's Hy account thanked Artificial Analysis for an evaluation that assigns Hy4 a score of 35 on its Intelligence Index v5 and places it second among models from China-based AI labs.
A November 5 report examines VAST's research decisions and support for young researchers, including separating world-state logic from generated visuals.
A November 5 hands-on report identifies the newer model in Xiaoyi Work, while noting that Huawei has not published detailed specifications.
Kimi's official account linked to model weights shortly after midnight Beijing time; a later technical summary describes four-bit distribution, with details requiring further verification.
