Qwen3.5-35B-A3B on Q-Bench Video
66.9%Rank #7 of 9 on Q-Bench Video
- Coverage
- Full (868/868)
- Cost per task
- —
- Median latency
- 4.5s
- Tokens per task
- 1,101 in · 1,265 out
Run setting: 1 fps.
Tokens and latency
Tokens per call
| Token type | Mean | Median | 90th pct. |
|---|---|---|---|
| Input | 1,101 | 1,055 | 2,117 |
| of which cached | 0 | 0 | 0 |
| Output | 1,265 | 651 | 2,888 |
Over 859 calls that returned an answer, including questions asked again after a failed attempt.
Latency
- Median
- 4.5s
- Mean
- 8.5s
- 90th percentile
- 18.8s
- 99th percentile
- 56.7s
We only consider the first question against a video for self hosted models, ignoring prefix caching latency.