MCSBench - L3: leaderboard
Measures multimodal conceptual-structure alignment on MCSBench, a diagnostic multiple-choice benchmark of 661 base visual question-answering items over fine-grained visual and taxonomy-like categories. L3 is the hypothetical reasoning level: predicting how conceptual relations shift under hypothetical edits. These are base-VQA accuracies from the released results table; the reasoning-chain integrity overlay, which carries the fourth metacognitive-verification level, is scored separately and is not part of this series. Five API models also have a run with web search enabled in that table; those runs are not part of this series either. This series reports MCSBench - L3 using Accuracy (%).
Metric: Accuracy (%). Source: huggingface.co. Saturation forecast: Around February 2027. 77 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Flash (Preview) | 74.81 | #90 |
| 2 | Gemini 3 Pro (Preview) | 72.52 | #71 |
| 3 | GPT-5 | 61.07 | #105 |
| 4 | Claude Opus 4.6 | 53.44 | #67 |
| 5 | Gemini 2.5 Pro | 50.38 | #159 |
| 6 | Gemini 2.5 Flash | 37.4 | #255 |
| 7 | GPT-5.2 | 36.64 | #115 |
| 8 | O4 Mini | 35.11 | #185 |
| 9 | GPT-5 Mini | 34.35 | #193 |
| 10 | Claude Sonnet 4.5 | 32.06 | #148 |
| 11 | Qwen 3 VL 30B A3B (Thinking) | 32.06 | #348 |
| 12 | Gemma 3 12B (IT) | 29.77 | #672 |
| 13 | Qwen 2.5 VL 72B Instruct | 28.24 | #383 |
| 14 | Qwen 3 VL 235B A22B Instruct | 28.24 | #280 |
| 15 | InternVL3-14B | 27.48 | #506 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=mcsbench-l3 · How It Works · Data refreshed daily, snapshot 2026-11-05.