Qwen3.8-27B on UCF101-AD
45.5%Rank #5 of 9 on UCF101-AD
- Coverage
- Full (4224/4224)
- Cost per task
- —
- Median latency
- 5.9s
- Tokens per task
- 1,054 in · 696 out
Run setting: 1 fps.
Tokens and latency
Tokens per call
| Token type | Mean | Median | 90th pct. |
|---|---|---|---|
| Input | 1,054 | 910 | 1,303 |
| of which cached | 0 | 0 | 0 |
| Output | 696 | 275 | 1,637 |
Over 4,224 calls that returned an answer, including questions asked again after a failed attempt.
Latency
- Median
- 5.9s
- Mean
- 14.5s
- 90th percentile
- 33.6s
- 99th percentile
- 127.3s
We only consider the first question against a video for self hosted models, ignoring prefix caching latency.