Gemini 4 Argon on LVBench
86.3%Rank #2 of 10 on LVBench
- Coverage
- 98.26% (1522/1549)
- Cost per task
- $0.591
- Median latency
- 91.8s
- Tokens per task
- 295,157 in · 35 out
Run setting: 1 fps, 5000-frame cap (34 videos thinned).
Tokens and latency
Tokens per call
| Token type | Mean | Median | 90th pct. |
|---|---|---|---|
| Input | 295,157 | 267,725 | 460,829 |
| of which cached | 45,329 | 0 | 221,719 |
| Output | 35 | 5 | 103 |
| of which reasoning | 30 | 0 | 98 |
Over 1,578 calls that returned an answer, including questions asked again after a failed attempt.
Latency
- Median
- 91.8s
- Mean
- 122.3s
- 90th percentile
- 229.3s
- 99th percentile
- 555.7s
Per call, end to end including file upload and processing.