Model Usage Ranking

To help you understand how the built-in models on the CloudBase AI platform are actually used, we update the share of token usage of each model periodically.

This issue covers September 21 – September 27, 2026 (Beijing Time), a full calendar week, and the metric is the total token volume of all text models on the platform (input + output tokens, including cache hits).

Token Usage DistributionShare of token volume by model · normalized to 100% per day

2026-09-21 ~ 2026-09-27

100%75%50%25%0%

09-2109-2209-2309-2409-2509-2609-27

7-day cumulative shareHover a bar for that day

1. 1DeepSeek-V4-Flash65.6%
2. 2Hunyuan Hy318.1%
3. 3GLM-5.34.5%
4. 4Hy4-preview3.9%
5. 5GLM-5.3-Flash2.7%
6. 6DeepSeek-V4-Pro-08131.8%
7. 7Other1.6%
8. 8Kimi-K31.2%
9. 9MiniMax-M30.7%

Each bar is normalized to 100% of that day's total; the right panel shows the 7-day cumulative share by default and switches to the hovered day's share. Values are merged from raw modelTags into model families.

Overview

Metric	Value
Period	2026-09-21 ~ 2026-09-27 (7 days)
Model families	8 (plus an Other bucket)
Top-end concentration	Top 2 model families account for 83.7%
Leading model	DeepSeek-V4-Flash, 65.56%