Qwen3.5-35B-A3B on EgoSchema
77.2%Rank #7 of 9 on EgoSchema
- Coverage
- Full (500/500)
- Cost per task
- —
- Median latency
- 13.4s
- Tokens per task
- 18,308 in · 2,096 out
Run setting: 1 fps.
Tokens and latency
Tokens per call
| Token type | Mean | Median | 90th pct. |
|---|---|---|---|
| Input | 18,308 | 18,283 | 18,452 |
| of which cached | 0 | 0 | 0 |
| Output | 2,096 | 1,251 | 4,383 |
Over 500 calls that returned an answer, including questions asked again after a failed attempt.
Latency
- Median
- 13.4s
- Mean
- 18.8s
- 90th percentile
- 33.8s
- 99th percentile
- 91.3s
We only consider the first question against a video for self hosted models, ignoring prefix caching latency.