# GroundingME: Multi-Dimensional Visual Grounding (GroundingME)

> A visual grounding benchmark of 1,005 examples that tests whether multimodal models localize objects across four dimensions: discriminative look-alikes, spatial relations, occluded or tiny targets, and ungroundable rejection queries.

Canonical page: https://benchlm.ai/benchmarks/groundingme

- Category: [Multimodal & Grounded](/multimodal-grounded)
- Last updated: GroundingME CVPR 2026 leaderboard (captured November 20, 2026)

## About GroundingME

- Year: 2025
- Tasks: 1,005 multi-dimensional visual grounding examples
- Format: Accuracy@0.5 (%)
- Difficulty: Fine-grained object localization and ungroundable rejection
- Paper: [GroundingME: Exposing the Visual Grounding Gap in MLLMs through Multi-Dimensional Evaluation](https://arxiv.org/abs/2512.17495)

Introduced by Li et al. (CVPR 2026) from Peking University and Xiaomi. We mirror the official CVPR 2026 leaderboard (32 models: 28 open-source and 4 commercial) with overall Accuracy@0.5 and the four sub-dimension scores (Discriminative, Spatial, Limited, Rejection). Provider-reported results for newer models not yet on the official board appear below under lab-reported results.

GroundingME is currently displayed on BenchLM for reference, but it is excluded from the weighted scoring formula.

## Leaderboard (32 models)

| Rank | Model | Configuration | Creator | Score |
|------|-------|---------------|---------|-------|
| 1 | [Qwen3-VL-A22B (Thinking)](https://groundingme.github.io/) | Open-source · Dis 65.2 · Spa 73.7 · Lim 45.0 · Rej 5.5 | Alibaba | 49.8% |
| 2 | [Qwen3-VL-32B (Thinking)](https://groundingme.github.io/) | Open-source · Dis 65.7 · Spa 70.0 · Lim 36.0 · Rej 9.5 | Alibaba | 46.9% |
| 3 | [Seed-1.6-Vision-250815 (Thinking)](https://groundingme.github.io/) | Commercial · Dis 59.3 · Spa 72.7 · Lim 41.7 · Rej 1.5 | ByteDance | 46.5% |
| 4 | [Qwen3-VL-235B-A22B Instruct](/models/qwen3-vl-235b-a22b-instruct) | Open-source · Dis 69.6 · Spa 49.7 · Lim 54.0 · Rej 0.0 | Alibaba | 45.1% |
| 5 | [Seed-1.6-Vision-250815](https://groundingme.github.io/) | Commercial · Dis 59.8 · Spa 58.7 · Lim 42.7 · Rej 1.0 | ByteDance | 42.6% |
| 6 | [Qwen3-VL-32B](https://groundingme.github.io/) | Open-source · Dis 75.0 · Spa 47.3 · Lim 34.0 · Rej 0.0 | Alibaba | 39.5% |
| 7 | [Qwen3-VL-A3B (Thinking)](https://groundingme.github.io/) | Open-source · Dis 53.4 · Spa 53.3 · Lim 38.0 · Rej 5.5 | Alibaba | 39.2% |
| 8 | [Qwen3-VL-A3B](https://groundingme.github.io/) | Open-source · Dis 63.2 · Spa 30.0 · Lim 46.7 · Rej 0.0 | Alibaba | 35.7% |
| 9 | [Qwen3-VL-8B (Thinking)](https://groundingme.github.io/) | Open-source · Dis 52.5 · Spa 43.0 · Lim 33.3 · Rej 4.5 | Alibaba | 34.3% |
| 10 | [GLM-4.5V (Thinking)](https://groundingme.github.io/) | Open-source · Dis 52.5 · Spa 45.3 · Lim 30.3 · Rej 4.0 | Z.ai | 34.0% |
| 11 | [Qwen3-VL-4B](https://groundingme.github.io/) | Open-source · Dis 56.4 · Spa 28.3 · Lim 47.0 · Rej 0.0 | Alibaba | 33.9% |
| 12 | [GLM-4.5V](https://groundingme.github.io/) | Open-source · Dis 52.9 · Spa 42.0 · Lim 29.3 · Rej 0.5 | Z.ai | 32.1% |
| 13 | [Qwen3-VL-8B](https://groundingme.github.io/) | Open-source · Dis 61.3 · Spa 26.3 · Lim 36.0 · Rej 0.0 | Alibaba | 31.0% |
| 14 | [Qwen2.5-VL-72B](https://groundingme.github.io/) | Open-source · Dis 48.5 · Spa 40.3 · Lim 23.7 · Rej 3.0 | Alibaba | 29.6% |
| 15 | [Qwen2.5-VL-32B](/models/qwen2-5-vl-32b) | Open-source · Dis 47.5 · Spa 40.0 · Lim 17.7 · Rej 0.0 | Alibaba | 26.9% |
| 16 | [MiMo-VL-7B-RL-2508 (Thinking)](https://groundingme.github.io/) | Open-source · Dis 46.6 · Spa 28.7 · Lim 17.0 · Rej 5.0 | Xiaomi | 24.1% |
| 17 | [Qwen3-VL-2B](https://groundingme.github.io/) | Open-source · Dis 44.6 · Spa 11.7 · Lim 28.7 · Rej 0.0 | Alibaba | 21.1% |
| 18 | [Gemini 2.5 Pro](/models/gemini-2-5-pro) | Commercial · Dis 34.8 · Spa 34.0 · Lim 7.0 · Rej 7.0 | Google | 20.7% |
| 19 | [Gemini 2.5 Flash](/models/gemini-2-5-flash) | Commercial · Dis 36.3 · Spa 25.0 · Lim 13.0 · Rej 0.0 | Google | 18.7% |
| 20 | [MiMo-VL-7B-RL-2508](https://groundingme.github.io/) | Open-source · Dis 44.1 · Spa 19.3 · Lim 13.0 · Rej 0.0 | Xiaomi | 18.6% |
| 21 | [InternVL3.5-A28B](https://groundingme.github.io/) | Open-source · Dis 28.4 · Spa 25.0 · Lim 13.0 · Rej 0.0 | OpenGVLab | 17.1% |
| 22 | [Qwen2.5-VL-7B](https://groundingme.github.io/) | Open-source · Dis 31.9 · Spa 14.3 · Lim 14.3 · Rej 0.5 | Alibaba | 15.1% |
| 23 | [Llama 4 Maverick](/models/llama-4-maverick) | Open-source · Dis 18.1 · Spa 22.3 · Lim 5.0 · Rej 6.0 | Meta | 13.0% |
| 24 | [Llama-Nemotron-8B](https://groundingme.github.io/) | Open-source · Dis 25.0 · Spa 6.0 · Lim 8.3 · Rej 5.5 | NVIDIA | 10.4% |
| 25 | [Llama 4 Scout](/models/llama-4-scout) | Open-source · Dis 17.6 · Spa 12.3 · Lim 3.7 · Rej 2.5 | Meta | 8.9% |
| 26 | [Keye-VL-1.5-8B](https://groundingme.github.io/) | Open-source · Dis 21.6 · Spa 8.0 · Lim 5.7 · Rej 0.0 | Kwai | 8.5% |
| 27 | [LLaVA-OneVision-1.5-8B](https://groundingme.github.io/) | Open-source · Dis 9.8 · Spa 4.7 · Lim 3.3 · Rej 0.0 | LLaVA | 4.4% |
| 28 | [MiniCPM-V-4.5](https://groundingme.github.io/) | Open-source · Dis 7.8 · Spa 4.0 · Lim 4.0 · Rej 0.0 | OpenBMB | 4.0% |
| 29 | [InternVL3.5-8B](https://groundingme.github.io/) | Open-source · Dis 6.4 · Spa 4.0 · Lim 1.7 · Rej 1.5 | OpenGVLab | 3.3% |
| 30 | [Mistral-3.2-24B](https://groundingme.github.io/) | Open-source · Dis 4.4 · Spa 2.7 · Lim 0.0 · Rej 0.0 | Mistral | 1.7% |
| 31 | [Gemma 3 27B](/models/gemma-3-27b) | Open-source · Dis 1.5 · Spa 0.3 · Lim 0.0 · Rej 0.0 | Google | 0.4% |
| 32 | [Phi-4 Multimodal Instruct](/models/phi-4-multimodal-instruct) | Open-source · Dis 1.0 · Spa 0.7 · Lim 0.0 · Rej 0.0 | Microsoft | 0.4% |

## Results reported by model developers

Unranked. Labs ran these themselves. A lab’s own result is shown for reference. It never takes a rank on this board.

| Model | Reported value | What was measured | Against the official board | Who ran it | Source |
|-------|----------------|-------------------|---------------------------|------------|--------|
| [Inkling 2 Mini](/models/inkling-2-mini) | 55.5% (Overall accuracy) | Reported in the Inkling 2 Mini model card; not yet evaluated on the official GroundingME CVPR 2026 leaderboard. | Comparability not established | Reported by Thinking Machines Lab | [Thinking Machines Lab Inkling 2 Mini model card](https://thinkingmachines.ai/news/inkling-2-mini/model-card/) |

## FAQ

### What does GroundingME measure?

A visual grounding benchmark of 1,005 examples that tests whether multimodal models localize objects across four dimensions: discriminative look-alikes, spatial relations, occluded or tiny targets, and ungroundable rejection queries.

### Which model leads the published GroundingME snapshot?

Qwen3-VL-A22B (Thinking) currently leads the published GroundingME snapshot with a score of 49.8%.

### How many models are evaluated on GroundingME?

The GroundingME CVPR 2026 leaderboard (captured November 20, 2026) contains 32 AI models.

### Does GroundingME affect BenchLM's overall score?

Not directly. GroundingME is still displayed on BenchLM for reference, but it is excluded from the weighted scoring formula.
