HAERAE-Vision - Shopping/Consumer: leaderboard

The 27 shopping and consumer questions of HAERAE-Vision (What Users Leave Unsaid, Choi et al., ACL 2026 Findings): real Korean community questions that ask to identify pictured products and brands or where to buy them, in the users' original under-specified wording. A GPT-5-mini judge scores each answer against a 1-5 item checklist (met 1, partly met 0.5, not met 0); the score is the mean checklist credit per question, in percent, averaged over three runs; higher is better.

Metric: Checklist Score (%; original under-specified queries, GPT-5-mini judge, mean of 3 runs). Source: arxiv.org. Saturation forecast: Around February 2027. 45 models tracked.

Top models

#ModelScoreOverall rank
1GPT-555.96#163
2Gemini 2.5 Flash46.13#452
3Gemini 2.5 Pro46#262
4GPT-5 Mini44.33#349
5Qwen 3 VL 32B Instruct43.38#582
6Qwen 3 VL 235B A22B Instruct40.98#521
7Qwen 3 VL 32B (Thinking)38.16#557
8Qwen 3 VL 30B A3B (Thinking)37.99#710
9Grok 436.73#306
10Qwen 3 VL 30B A3B Instruct36.53#870
11Qwen 3 VL 235B A22B (Thinking)34.18#451
12Qwen 3 VL 8B (Thinking)32.53#887
13Qwen 3 VL 4B (Thinking)31.29#1303
14Gemini 2.5 Flash Lite28.82#958
15Qwen 3 VL 8B Instruct27.98#982

No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.

Interactive version: theaggregate.ai/benchmark?slug=haerae-vision-shopping-consumer · How It Works · Data refreshed daily, snapshot 2026-11-12.