SYNCR: leaderboard

SYNCR (2026) tests cross-video reasoning in multimodal models with synthetic videos from Habitat, Kubric and CLEVRER: temporal alignment, spatial tracking, comparative reasoning and holistic synthesis across 2 to 4 videos. The score is the mean exact-match accuracy (%) over its eight tasks, on 200 four-option questions per task; higher is better.

Metric: Average exact-match accuracy (%) over the eight SYNCR tasks, zero-shot four-option multiple choice over 2 to 4 labelled input videos, exact match, deterministic decoding, on the paper's balanced subset of 200 questions per task; chance is 25; higher is better. Source: arxiv.org. Saturation forecast: Around July 2027. 18 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Gemini 3 Flash52.5#175
2GPT-5.450.5#135
3Qwen 3 VL 32B Instruct47.3#606
4Qwen 3 VL 8B Instruct42.4#1029
5LLaVA-OV-72B41.7#1879
6Qwen 2.5 VL 72B Instruct40.6#841
7InternVL3.5-38B38.6#1021
8Qwen 2.5 VL 32B Instruct36.7#1062
9Qwen 3 VL 4B Instruct35.6#1332
10Qwen 2.5 VL 7B Instruct35.1#1902
11InternVL3.5-14B34.1#1504
12LLaVA-OV-7B33.5#4475
13InternVL3.5-4B31.6#1930
14LLaVA-OV-0.5B30.5#4352
15Qwen 3 VL 2B Instruct30.3#2367

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

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