Kimi-VL-16B-A3B: benchmark results
Provider: Moonshot. Access: API.
Aggregate Rating 1583 ± 36, rank #1205 of 10916 rated models, from 17 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| CourtSI-Bench - Ball Counting | 46.4 | Exact-match accuracy (%) on the 28 ball-counting questions, | 70.8 |
| CourtSI-Bench - Height | 56.4 | Threshold mean relative accuracy (%), averaged over confiden | 63.6 |
| CourtSI-Bench - Ball-Zone | 56.5 | Exact-match accuracy (%) on the 255 ball-to-court-zone relat | 56.5 |
| Video-MME-v2 - Level 1 Retrieval & Aggregation (w/o subs) | 16.39 | Level 1 Score w/o sub (%) | 54.2 |
| CourtSI-Bench - Ball-Player | 57.6 | Exact-match accuracy (%) on the 297 ball-to-player relation | 47.7 |
| CourtSI-Bench - Object-Line Distance | 19.5 | Threshold mean relative accuracy (%), averaged over confiden | 40.9 |
| CourtSI-Bench - Object-Object Distance | 16.7 | Threshold mean relative accuracy (%), averaged over confiden | 40.9 |
| CourtSI-Bench | 34.7 | Overall score (%): the question-weighted mean of the 13 task | 38.6 |
| Video-MME-v2 - Non-Lin Score (w/o subs) | 11.7 | Non-Lin Score w/o sub (%) | 37.5 |
| Video-MME-v2 | 27.3 | Avg Accuracy w/o sub (%) | 36.5 |
| CourtSI-Bench - Player-Zone | 32.9 | Exact-match accuracy (%) on the 82 player-to-court-zone rela | 31.2 |
| Video-MME-v2 - Level 3 Complex Reasoning (w/o subs) | 9.73 | Level 3 Score w/o sub (%) | 29.2 |
Interactive version: theaggregate.ai/model?slug=kimi-vl-16b-a3b · How It Works · Data refreshed daily, snapshot 2026-11-08.