Gemini 3.8 Flash on LVBench
86.4%Rank #1 of 9 on LVBench
- Coverage
- Full (1549/1549)
- Cost per task
- $0.217
- Median latency
- 67.1s
- Tokens per task
- 286,019 in · 658 out
Run setting: 1 fps, 5889-frame cap (18 videos thinned).
Tokens and latency
Tokens per call
| Token type | Mean | Median | 90th pct. |
|---|---|---|---|
| Input | 286,019 | 291,446 | 405,464 |
| of which cached | 285,970 | 291,410 | 405,431 |
| Output | 658 | 326 | 1,420 |
| of which reasoning | 656 | 324 | 1,419 |
Over 1,549 calls that returned an answer, including questions asked again after a failed attempt.
Latency
- Median
- 67.1s
- Mean
- 85.7s
- 90th percentile
- 154.9s
- 99th percentile
- 382.8s
Per call, end to end including file upload and processing.