Physix FrontierPhysix Frontier · News Briefing Card (QbitAI · Oct 12, 2026)
Zhipu GLM-5.3-Highspeed Hits 490 tokens/s in Zcode Gray Test
KEY FACTS
- Zhipu has launched the GLM-5.3-Highspeed model, now in random gray release to subscribers on Zcode 3.15.1.
- A user measured the model's output speed at 490 tokens/s.
- The model has a context window of 1 million tokens and a maximum generation limit of about 128k.
- The model offers three thinking intensity settings: low, high, and max.
- Zhipu's GLM-5.1-highspeed, released in May this year, ran at 400 tokens/s.
KEY DATA
490 tokens/sMeasured output speed
1 million tokensContext window
131,072Max generation tokens
400 tokens/sGLM-5.1-highspeed speed
PHYSIX OBSERVATION
If 490 tokens/s holds up, it will set a new speed ceiling for domestic large-model APIs, with long-horizon tasks and agent scenarios benefiting most directly. But the figure comes from a single user measurement, and the model is only in random gray release, so real-world stable throughput and pricing have not been announced. Zhipu's repeated use of high-speed versions to play the speed card shows that inference efficiency is becoming a new focal point of large-model competition.
Source: QbitAI report