AICA-Bench - Emotion Understanding (CoT): leaderboard

AICA-Bench emotion understanding with chain-of-thought prompts that reason over color, scene and facial cues before classifying the emotion of an image. Weighted F1 in percent, higher is better.

Metric: Weighted F1 (%) of emotion classification (expressed emotion recognition and evoked emotion prediction) with chain-of-thought prompts that reason over color, scene and facial cues first, over AICA-Bench (8,086 affective images from nine public emotion datasets with GPT-4o-generated instructions; closed-source models through their APIs at standard settings, open models on A100 GPUs); higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 23 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Gemini 2.5 Flash69.32#463
2Gemini 2.0 Flash68.98#723
3Qwen-VL-Plus (AICA-Bench checkpoint unspecified)67.81
4Gemini 2.5 Pro67.57#276
5Qwen-VL-Max (AICA-Bench checkpoint unspecified)65.98
6GPT-4o65.42#751
7GPT-4o Mini63.68#1667
8Qwen 2.5 VL 7B Instruct57.25#1902
9Qwen 2 VL 7B Instruct55.19#2573
10Ovis2-16B54.7#1594
11Qwen 2.5 VL 3B Instruct53.67#2713
12LLaVA-OneVision53.25#2028
13InternVL3-8B-Instruct52.98#2872
14Ovis2-8B52.73#2047
15InternVL3-14B-Instruct52.04#1801

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

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