FGTR-Bench: leaderboard
FGTR-Bench (Fine-Grained Traffic Reasoning Benchmark, 2026) tests whether multimodal models find small, localized evidence in traffic scenes through 4-option multiple-choice questions. The board is overall accuracy in percent on the 4,947-question blind test split for models without FGTR-Bench training; higher is better.
Metric: Overall accuracy (%) on the 4,947-question blind test split, 4-option multiple-choice questions on the blind test split, zero-shot (models without FGTR-Bench training), greedy letter decoding; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 5 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 2.5 VL 7B Instruct | 69.9 | #1833 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=fgtr-bench · How It Works · Data refreshed daily, snapshot 2026-11-13.