Human-Grounded Multimodal Benchmark with 900K: leaderboard
Authentic school examinations provide a high-validity test bed for evaluating multimodal large language models (MLLMs), yet benchmarks grounded in Japanese K-12 assessments remain scarce. We present a multimodal dataset constructed from Japan's National Assessment of Academic Ability, comprising officially released middle-school items in Science, Mathematics, and Japanese Language. Unlike existing
Metric: Science MC Accuracy (self-reported). Source: benchmarklist.com. Saturation forecast: Around May 2027. 11 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 | 90.9 | #98 |
| 2 | Gemini 2.5 Pro | 72.7 | #152 |
| 3 | Gemini 2.5 Flash | 72.7 | #245 |
| 4 | Claude Sonnet 4 | 72.7 | #199 |
| 5 | Gemini 3.1 Pro (Preview) | 63.6 | #58 |
| 6 | Claude Sonnet 4.5 | 63.6 | #142 |
| 7 | Gemma 3 12B (IT) | 54.5 | #664 |
| 8 | Claude 3.7 Sonnet | 54.5 | #249 |
| 9 | GPT-4o | 45.5 | #343 |
| 10 | GPT-4o Mini | 45.5 | #599 |
| 11 | Qwen 2.5 VL 7B Instruct | 36.3 | #654 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=human-grounded-multimodal-benchmark-with-900k · How It Works · Data refreshed daily, snapshot 2026-10-20.