Event ·Official first-hand

智谱发布GLM-5.3-FlashX，推理速度达200 tokens/s

智谱AI推出的GLM-5.3-FlashX模型，在约10万颗国产加速器上实现每秒200个token的推理速度。该模型基于320亿参数的混合专家（MoE）架构，并使用基础设施代理优化了服务堆栈。

Current read

智谱AI新出的GLM-5.3-FlashX模型，在国产加速器上跑得特别快，每秒能处理200个token，比之前版本效率高不少。

10 sources78° AI heatLast updated 2026/9/18 07:51:45

Evidence

10 sources

Primary sourcepandaily

2026/9/18 07:51:45View source

Related source 1小互

2026/9/17 02:10:49View source

Related source 2Browser Use

2026/9/19 19:22:51View source

Related source 3shao__meng

2026/9/16 13:10:05View source

Related source 4Genk

2026/9/16 15:24:00View source

Related source 5Fireworks AI

2026/9/16 22:09:49View source

Related source 6量子位

2026/9/19 11:43:12View source

Related source 7Tech in Asia

2026/9/17 00:50:24View source

Related source 8Yangyi

2026/9/18 10:44:26View source

Related source 9CNBC Indonesia Tech

2026/9/19 06:45:00View source

Conflict check

Dedup data only shows that these sources discuss the same event, not that they agree. There is no structured supporting or refuting evidence yet, so no conflict is inferred automatically.

Open the story