IMEB (Direct Answer): leaderboard
IMEB, the Image Multi-Entity Benchmark introduced with the HyperEyes search agent (Li et al., 2026), has 300 human-curated questions about images of several entities (4.6 on average) that require locating and looking up more than one of them. Accuracy in percent, judged by an LLM against the reference answer; higher is better. On this board models answer without tools, from parametric knowledge alone.
Metric: Accuracy (%; LLM-judged against the reference answer, 300 human-curated multi-entity image questions, answered without tools). Source: arxiv.org. Saturation forecast: Around February 2027. 5 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 40.8 | #91 |
| 2 | Kimi K2.5 | 27.7 | #252 |
| 3 | Claude Opus 4.6 | 27 | #100 |
| 4 | Qwen 3 VL 235B A22B Instruct | 12 | #521 |
| 5 | Qwen 3 VL 30B A3B Instruct | 6.7 | #870 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=imeb-direct-answer · How It Works · Data refreshed daily, snapshot 2026-11-12.