Qwen3.8-Max on LVBench
58.9%Rank #7 of 9 on LVBench
- Coverage
- 70.82% (1097/1549)
- Cost per task
- $0.542
- Median latency
- 131.5s
- Tokens per task
- 259,870 in · 3,647 out
Run setting: 1 fps native video, ~256k visual-token ceiling (all videos thinned to ~860 frames).
Tokens and latency
Tokens per call
| Token type | Mean | Median | 90th pct. |
|---|---|---|---|
| Input | 259,870 | 260,815 | 275,538 |
| of which cached | 242,103 | 258,048 | 274,432 |
| of which frames or video | 259,773 | 260,733 | 275,457 |
| Output | 3,647 | 2,631 | 8,779 |
Over 1,295 calls that returned an answer, including questions asked again after a failed attempt.
Latency
- Median
- 131.5s
- Mean
- 147.3s
- 90th percentile
- 266.0s
- 99th percentile
- 386.0s
Per call, end to end including file upload and processing.