OmniCoT - See: leaderboard

Metric: Accuracy (%) on See questions (multi-step orientation tracking, relative angular calculation) with chain-of-thought prompting on OmniCoT-B (synthetic 360-degree equirectangular panoramas with three anchor objects given by coordinates and orientation); answers are object identities, visibility, directions or the closest angle option, exact match after normalization. Source: arxiv.org. Saturation forecast: Around March 2028. 21 models tracked.

Top models

#ModelScore
1GPT-557.29
2Seed 1.857.09
3Grok 454.4
4GLM-4.5V53.15
5Qwen 2.5 VL 72B Instruct52.03
6ChatGPT 4o51.97
7Qwen 2.5 VL 32B Instruct47.84
8Gemini 3 Flash47.52
9Qwen 3 VL 8B Instruct41.52
10Qwen 2.5 VL 7B Instruct33.77

Interactive version: theaggregate.ai/benchmark?slug=omnicot-see · How It Works · Data refreshed daily, snapshot 2026-09-29.