GLM-5.3 HighSpeed
Zhipu began a gray-release test of GLM-5.3 HighSpeed in its ZCode coding tool, with users reporting output above 490 tokens per second.
GLM-5.3 HighSpeed has the same 1 million token context window and 131,072 token maximum output as GLM-5.3, so the difference is inference speed. Reported speed is above 490 tokens per second, against 70.84 for the standard GLM-5.3 API and 121.34 for DeepSeek V4.1 Flash, per AI Ping. An earlier GLM-5.1-highspeed API reached 400 tokens per second, built with the TileRT team, which compiles the inference graph statically.
- Date
- Monday 12 October 2026
- Lab
- Zhipu AI / Z.ai
- Kind
- model
- Access
- app only
Figures
| Measure | Value | Measured by |
|---|---|---|
| Output speed | above 490 tokens/s seen by one tester in gray release; may not hold at scale | Leo (X user), reported by ZheDongXi |
| GLM-5.3 standard API output speed | 70.84 tokens/s comparison figure | AI Ping |
| DeepSeek V4.1 Flash official API output speed | 121.34 tokens/s comparison figure | AI Ping |
| Context window | 1,000,000 tokens max output 131,072 tokens | Pi harness catalog, via press |
Available only to a random subset of ZCode subscribers, with no official announcement, launch date or pricing. The speed figure comes from a single user's screenshot.
Sources
This record was checked against its sources on 12 October 2026. How we check