Nemotron 3 Nano Omni 30B-A3B on LVBench
54.2%Rank #8 of 9 on LVBench
- Coverage
- Full (1549/1549)
- Cost per task
- —
- Median latency
- 32.8s
- Tokens per task
- 104,641 in · 6,963 out
Run setting: 1 fps, 727-frame cap (103 videos thinned).
Tokens and latency
Tokens per call
| Token type | Mean | Median | 90th pct. |
|---|---|---|---|
| Input | 104,641 | 104,701 | 104,732 |
| of which cached | 0 | 0 | 0 |
| Output | 6,963 | 1,987 | 17,414 |
Over 1,549 calls that returned an answer, including questions asked again after a failed attempt.
Latency
- Median
- 32.8s
- Mean
- 58.8s
- 90th percentile
- 129.7s
- 99th percentile
- 141.8s
We only consider the first question against a video for self hosted models, ignoring prefix caching latency.