FCMBench-Video (English) - Counting: leaderboard
FCMBench-Video (2026) English-language handheld document videos from financial credit workflows, task counting: exact-match accuracy (%) of the de-duplicated count of logical documents shown in the video. Zero-shot; higher is better.
Metric: Counting: exact-match accuracy (%) of the de-duplicated count of logical documents shown in the video, on FCMBench-Video's en-US subset (795 composed videos of 8 English document types, 5,362 instances); zero-shot with structured outputs; each model's native video serving path (2 FPS sampling for most open models, 0.5 FPS for Ovis2.5, raw video upload for the API models); higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 9 models tracked. Updated 7 Oct 2026.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Pro (Preview) | 67.92 | #119 |
| 2 | Doubao-Seed-1.6-vision | 67.77 | #361 |
| 3 | Qwen 3.5 27B | 59.77 | #375 |
| 4 | Qwen 3 VL 32B Instruct | 54.94 | #606 |
| 5 | InternVL3-8B | 38.42 | #1782 |
| 6 | Qwen3 Omni 30B A3B Instruct | 34.72 | #930 |
| 7 | Qwen 3 VL 8B Instruct | 34.57 | #1029 |
| 8 | Kimi-VL-A3B-Instruct | 30.11 | #2253 |
| 9 | Ovis2.5-9B | 27.25 | #1225 |
No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.
Interactive version: theaggregate.ai/benchmark?slug=fcmbench-video-english-counting · How It Works · Data refreshed daily, snapshot 2026-11-21.