V2X-QA - Ego Risk: leaderboard

V2X-QA (2026) asks multimodal models four-option driving questions about real vehicle-mounted and roadside camera views of the same scenes, answered with one option letter on the expert-verified test split. This board is the vehicle-side task on localized safety risks. Accuracy in percent; higher is better.

Metric: Accuracy (%) on the vehicle-side Ego Risk task (VS3); four-option multiple-choice questions on the expert-verified test split, answered as a single option letter under the paper's unified prompt; higher is better. Source: arxiv.org. Saturation forecast: Not forecast. 10 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Qwen 2.5 VL 72B Instruct51.3#841
2Qwen 3.5 Plus44.5#242
3Intern-S142.3#624
4Gemini 2.5 Pro40.4#276
5Qwen3-VL-Flash39#728
6GPT-5 Mini (2025-08-07)32.1#345
7Intern-S1-mini29.8#1178
8Qwen 3 VL 30B A3B Instruct27.5#895
9GPT-5.226#185
10Gemini 3 Flash (Preview)20.8#152

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=v2x-qa-ego-risk · How It Works · Data refreshed daily, snapshot 2026-11-21.