Z AI的GLM-5.2开源，百万token上下文，数学和编程全面超过GPT-5.5，值得试玩。
Z AI发布了GLM-5.2，一款开源权重模型，支持1M token上下文窗口。在long-horizon coding基准上得分为74.4，超过GPT-5.5的72.6。在SWE-bench Pro上得分为62.1，同样领先GPT-5.5。AIME 2026数学测试得分为99.2，高于Opus 4.8和GPT-5.5。该模型在Designarena排名第一，并以MIT许可证发布。
Chinese lab Z AI just released GLM-5.2, an impressive new open weights model with a 1M token context...
Chinese lab Z AI just released GLM-5.2, an impressive new open weights model with a 1M token context window. A few benchmark comparisons, slotting between Opus 4.8 and GPT 5.5 on several fronts: - 74.4 on long-horizon coding, ahead of GPT-5.5's 72.6. - 62.1 on SWE-bench Pro, ahead of GPT-5.5 again. - 99.2 on the AIME 2026 math set, ahead of both Opus 4.8 and GPT-5.5. GLM-5.2 also jumped to No. 1 in the @Designarena , overtaking Fable(!). @Zai_org shipped the weights under an MIT license, with its launch post calling it "technical access without borders." 💬 3 🔄 1 ❤️ 10 👀 1023 📊 4 ⚡