Video-Arena - Visual gfootball: leaderboard
Video-Arena: 180 cases, 20 per game. Strict tuple F1 requires every non-temporal field to match and timestamps within 3 seconds; overall is the mean of nine game means. Paper v1 Appendix B.3 fixes high effort for OpenAI, Anthropic and Gemini; Kimi/MiniMax use unspecified provider defaults. Qwen thinking modes remain distinct. One request, 1 FPS capped at 256 uniform frames; no tools. Metric: Strict tuple F1 (%).
Metric: Strict tuple F1 (%). Source: videoarena-bench.github.io. Saturation forecast: Around 2033. 10 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Kimi K3 | 0.8 | #33 |
| 2 | GPT-5.6 Sol (High) | 0.8 | #19 (GPT-5.6 Sol) |
| 3 | Claude Opus 5 (High) | 0.8 | #13 (Claude Opus 5) |
| 4 | Qwen 3.5 122B A10B | 0.3 | #179 |
| 5 | MiniMax-M3 | 0.2 | #150 |
| 6 | Gemini 3.7 Flash (High) | 0 | #22 (Gemini 3.7 Flash) |
| 7 | GPT-5.6 Luna (High) | 0 | #76 (GPT-5.6 Luna) |
| 8 | Qwen 3.8 27B (Non-reasoning) | 0 | #112 (Qwen 3.8 27B) |
No result here: #5 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=video-arena-visual-gfootball · How It Works · Data refreshed daily, snapshot 2026-10-26.