Molmo 2 8B on LVBench
48.6%Rank #9 of 9 on LVBench
- Coverage
- Full (1549/1549)
- Cost per task
- —
- Median latency
- 198.7s
- Tokens per task
- 30,385 in · 2 out
Run setting: 1 fps, 338-frame cap (103 videos thinned).
Tokens and latency
Tokens per call
| Token type | Mean | Median | 90th pct. |
|---|---|---|---|
| Input | 30,385 | 30,377 | 30,405 |
| of which cached | 28,309 | 30,320 | 30,336 |
| Output | 2 | 2 | 2 |
Over 1,549 calls that returned an answer, including questions asked again after a failed attempt.
Latency
- Median
- 198.7s
- Mean
- 198.9s
- 90th percentile
- 199.1s
- 99th percentile
- 200.7s
We only consider the first question against a video for self hosted models, ignoring prefix caching latency.