Nemotron 3 Nano Omni 30B-A3B on NExT-QA
81.2%Rank #9 of 9 on NExT-QA
- Coverage
- Full (4996/4996)
- Cost per task
- —
- Median latency
- 4.2s
- Tokens per task
- 6,310 in · 1,357 out
Run setting: 1 fps.
Tokens and latency
Tokens per call
| Token type | Mean | Median | 90th pct. |
|---|---|---|---|
| Input | 6,310 | 5,204 | 12,579 |
| of which cached | 0 | 0 | 0 |
| Output | 1,357 | 479 | 2,908 |
Over 5,024 calls that returned an answer, including questions asked again after a failed attempt.
Latency
- Median
- 4.2s
- Mean
- 8.7s
- 90th percentile
- 16.2s
- 99th percentile
- 102.7s
We only consider the first question against a video for self hosted models, ignoring prefix caching latency.