StemBind: leaderboard
Multimodal large language models (MLLMs) often know the rule but pick the wrong answer: on abstract visual reasoning (AVR) tasks, a model can describe what it sees and name the underlying pattern, yet still fail to choose the matching candidate. Existing AVR benchmarks cannot detect this because they collapse perception, rule induction, and answer selection into a single right-or-wrong signal. We
Metric: F Overall (self-reported). Source: benchmarklist.com. Saturation forecast: Not forecast. 24 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3.5 Plus | 42.2 | #125 |
| 2 | Qwen 3.5 27B | 41.3 | #187 |
| 3 | Qwen 3.5 35B A3B | 39.3 | #256 |
| 4 | Qwen 3.5 9B | 37.4 | #379 |
| 5 | Kimi K2.5 | 36.8 | #144 |
| 6 | Gemini 3.1 Pro (Preview) | 35.9 | #58 |
| 7 | Qwen 3.5 2B | 35.9 | #1148 |
| 8 | GLM-4.5V | 34.6 | #346 |
| 9 | Qwen 3.5 122B A10B | 33.8 | #173 |
| 10 | Qwen 3.5 4B | 33.8 | #475 |
| 11 | Gemma 4 31B (IT) | 33.1 | #176 |
| 12 | Gemma 4 26B A4B (IT) | 33 | #297 |
| 13 | GPT-5.4 | 31.8 | #79 |
| 14 | Claude Opus 4.7 | 30.7 | #49 |
| 15 | gemma-4-E4B-it | 30.6 | #566 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=stembind · How It Works · Data refreshed daily, snapshot 2026-10-20.