Nemotron 3 Nano Omni 30B-A3B on Video-MME v2
35.0%Rank #8 of 9 on Video-MME v2
- Coverage
- Full (3200/3200)
- Cost per task
- —
- Median latency
- 40.2s
- Tokens per task
- 78,696 in · 8,466 out
Run setting: 1 fps, 727-frame cap (254 videos thinned).
Tokens and latency
Tokens per call
| Token type | Mean | Median | 90th pct. |
|---|---|---|---|
| Input | 78,696 | 80,677 | 104,887 |
| of which cached | 0 | 0 | 0 |
| Output | 8,466 | 5,205 | 17,411 |
Over 3,344 calls that returned an answer, including questions asked again after a failed attempt.
Latency
- Median
- 40.2s
- Mean
- 61.8s
- 90th percentile
- 129.4s
- 99th percentile
- 143.6s
We only consider the first question against a video for self hosted models, ignoring prefix caching latency.