Nemotron 3 Nano Omni 30B-A3B on Q-Bench Video
64.9%Rank #8 of 9 on Q-Bench Video
- Coverage
- Full (868/868)
- Cost per task
- —
- Median latency
- 1.2s
- Tokens per task
- 1,525 in · 584 out
Run setting: 1 fps.
Tokens and latency
Tokens per call
| Token type | Mean | Median | 90th pct. |
|---|---|---|---|
| Input | 1,525 | 1,485 | 2,952 |
| of which cached | 0 | 0 | 0 |
| Output | 584 | 167 | 868 |
Over 868 calls that returned an answer, including questions asked again after a failed attempt.
Latency
- Median
- 1.2s
- Mean
- 3.4s
- 90th percentile
- 4.5s
- 99th percentile
- 95.5s
We only consider the first question against a video for self hosted models, ignoring prefix caching latency.