Qwen3.5-35B-A3B on LVBench
64.4%Rank #6 of 9 on LVBench
- Coverage
- Full (1549/1549)
- Cost per task
- —
- Median latency
- 23.3s
- Tokens per task
- 76,952 in · 4,688 out
Run setting: 1 fps, 768-frame cap (103 videos thinned).
Tokens and latency
Tokens per call
| Token type | Mean | Median | 90th pct. |
|---|---|---|---|
| Input | 76,952 | 76,821 | 76,897 |
| of which cached | 0 | 0 | 0 |
| Output | 4,688 | 1,674 | 13,373 |
Over 1,549 calls that returned an answer, including questions asked again after a failed attempt.
Latency
- Median
- 23.3s
- Mean
- 43.0s
- 90th percentile
- 102.7s
- 99th percentile
- 226.1s
Per call, end to end including file upload and processing.