关于“哪个 AI 模型最聪明”的竞争已经结束

📄 中文摘要

2026年1月至2月间,AI领域发生了前所未有的事件,十个主要的开放权重大型语言模型(LLM)架构在短短八周内相继发布。这些模型包括GLM-5、Kimi K2.5、Step 3.5 Flash等,各自的参数规模和性能水平各不相同。例如,GLM-5的总参数达到7440亿,活跃参数为400亿,其性能与GPT-5.2和Claude Opus 4.6相匹配。Kimi K2.5则以1万亿的参数量在发布时被评为前沿级别。这一系列发布标志着AI模型竞争的一个新阶段,可能会改变未来的技术格局。

📄 English Summary

The Competition Over "Which AI Model Is Smartest" Is Over.

Between January and February 2026, an unprecedented event occurred in the AI landscape with the public release of ten major open-weight large language model (LLM) architectures within just eight weeks. These models, including GLM-5, Kimi K2.5, and Step 3.5 Flash, vary significantly in terms of parameter counts and performance levels. For instance, GLM-5 boasts a total of 744 billion parameters with 40 billion active parameters, matching the performance of GPT-5.2 and Claude Opus 4.6. Kimi K2.5, with a staggering 1 trillion parameters, was classified as frontier-class at its release. This series of releases marks a new phase in the competition among AI models, potentially reshaping the future technological landscape.

Powered by Cloudflare Workers + Payload CMS + Claude 3.5

数据源: OpenAI, Google AI, DeepMind, AWS ML Blog, HuggingFace 等