Qwen3.8-Max on Video-MME v2
60.2%Rank #4 of 9 on Video-MME v2
- Coverage
- 98.5% (3152/3200)
- Cost per task
- $0.370
- Median latency
- 145.7s
- Tokens per task
- 166,837 in · 6,060 out
Run setting: 1 fps native video, ~256k visual-token ceiling (168 videos thinned).
Tokens and latency
Tokens per call
| Token type | Mean | Median | 90th pct. |
|---|---|---|---|
| Input | 166,837 | 161,564 | 255,407 |
| of which cached | 127,655 | 131,072 | 252,928 |
| of which frames or video | 166,614 | 161,213 | 255,265 |
| Output | 6,060 | 5,912 | 11,492 |
Over 3,565 calls that returned an answer, including questions asked again after a failed attempt.
Latency
- Median
- 145.7s
- Mean
- 152.3s
- 90th percentile
- 262.2s
- 99th percentile
- 363.6s
Per call, end to end including file upload and processing.