Kimi K3 on NExT-QA
86.3%Rank #6 of 9 on NExT-QA
- Coverage
- Full (4996/4996)
- Cost per task
- $0.0411
- Median latency
- 8.7s
- Tokens per task
- 12,100 in · 323 out
Run setting: 1 fps.
Tokens and latency
Tokens per call
| Token type | Mean | Median | 90th pct. |
|---|---|---|---|
| Input | 12,100 | 10,028 | 24,340 |
| of which cached | 10,585 | 8,960 | 23,552 |
| of which frames or video | 11,724 | 9,684 | 23,763 |
| Output | 323 | 134 | 527 |
| of which reasoning | 307 | 118 | 509 |
Over 5,030 calls that returned an answer, including questions asked again after a failed attempt.
Latency
- Median
- 8.7s
- Mean
- 17.2s
- 90th percentile
- 27.0s
- 99th percentile
- 167.2s
Per call, end to end including file upload and processing.