Qwen3.5-35B-A3B on Video-MME v2
42.5%Rank #7 of 9 on Video-MME v2
- Coverage
- Full (3200/3200)
- Cost per task
- —
- Median latency
- 37.6s
- Tokens per task
- 56,175 in · 6,105 out
Run setting: 1 fps, 768-frame cap (226 videos thinned).
Tokens and latency
Tokens per call
| Token type | Mean | Median | 90th pct. |
|---|---|---|---|
| Input | 56,175 | 56,222 | 77,012 |
| of which cached | 0 | 0 | 0 |
| Output | 6,105 | 3,786 | 15,152 |
Over 3,452 calls that returned an answer, including questions asked again after a failed attempt.
Latency
- Median
- 37.6s
- Mean
- 52.6s
- 90th percentile
- 112.4s
- 99th percentile
- 195.0s
We only consider the first question against a video for self hosted models, ignoring prefix caching latency.