Qwen3.8-Max 碾壓 Anthropic — 阿里將開源權重
阿里發布了參數量 2.4 萬億的 qwen3.8-max，其基準測試表現超越 claude fable 5，並計畫下週開源權重。此舉加劇了與美國實驗室的競爭，並挑戰了閉源前沿 ai 的主導地位。
alibaba launched qwen3.8-max, its largest model to date with 2.4 trillion parameters, asserting it rivals or surpasses top us systems like anthropic's claude fable 5 and openai's models across multiple benchmarks. internal tests and rankings on arena.ai place it competitively in text, coding, and visual tasks, trailing only specific fable and opus variants. the company will release the model's weights next week, signaling a strategic return to open-weight distribution after earlier proprietary moves. this challenges the closed-source models of us frontier labs and aligns with beijing's push for open-weight adoption to expand global influence. the release narrows the performance gap between chinese and american ai, significantly impacting the open-source ecosystem and geopolitical tech landscape.