GPT-6 Astra on Video-MME v2
73.4%Rank #1 of 9 on Video-MME v2
- Coverage
- Full (3200/3200)
- Cost per task
- $1.567
- Median latency
- 15.0s
- Tokens per task
- 156,195 in · 95 out
Run setting: 1 fps, 1500-frame cap (2 videos thinned).
Tokens and latency
Tokens per call
| Token type | Mean | Median | 90th pct. |
|---|---|---|---|
| Input | 156,195 | 141,039 | 258,081 |
| of which cached | 117,732 | 119,002 | 247,765 |
| Output | 95 | 45 | 235 |
| of which reasoning | 89 | 38 | 228 |
Over 3,217 calls that returned an answer, including questions asked again after a failed attempt.
Latency
- Median
- 15.0s
- Mean
- 16.6s
- 90th percentile
- 26.4s
- 99th percentile
- 41.5s
Per call, end to end including file upload and processing.