Qwen3.8-Max on NExT-QA
87.6%Rank #4 of 9 on NExT-QA
- Coverage
- 98.58% (4925/4996)
- Cost per task
- $0.0232
- Median latency
- 18.2s
- Tokens per task
- 5,372 in · 2,083 out
Run setting: 1 fps native video.
Tokens and latency
Tokens per call
| Token type | Mean | Median | 90th pct. |
|---|---|---|---|
| Input | 5,372 | 3,990 | 10,407 |
| of which cached | 4,512 | 3,456 | 10,240 |
| of which frames or video | 5,281 | 3,888 | 10,300 |
| Output | 2,083 | 608 | 6,453 |
Over 6,168 calls that returned an answer, including questions asked again after a failed attempt.
Latency
- Median
- 18.2s
- Mean
- 49.7s
- 90th percentile
- 143.1s
- 99th percentile
- 251.0s
Per call, end to end including file upload and processing.