# GroundingME leaderboard

external benchmark

> Every figure below is reproduced as its upstream source published it: nothing is modelled, estimated, interpolated or converted. `Unavailable` means no source published the value — it is never a zero. Each value carries its evidence state and the date it was observed.

Page: https://modelscale.dev/leaderboards/benchmarks/external-grounding-me

GroundingME: Multi-Dimensional Visual Grounding. Every model the catalog carries a published GroundingME value for, ranked by that value.

## Benchmark record

| Field | Value |
| --- | --- |
| Category | external |
| Measure | Accuracy@0.5 (%) |
| Tasks | 1,005 multi-dimensional visual grounding examples |
| Difficulty | Fine-grained object localization and ungroundable rejection |
| Published by | [GroundingME: Exposing the Visual Grounding Gap in MLLMs through Multi-Dimensional Evaluation](https://arxiv.org/abs/2512.17495) |

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.

## GroundingME ranking

8 models with a published GroundingME value, ordered by that value, highest first. Models the source has not scored on this benchmark are not listed — they are unmeasured, not last.

| Rank | Model | Provider | Accuracy@0.5 (%) | Evidence |
| --- | --- | --- | --- | --- |
| 1 | [Qwen3-VL-235B-A22B Instruct](https://modelscale.dev/models/qwen3-vl-235b-a22b-instruct) | Alibaba | 45.1 | Observed 2026-11-21 · source benchlm:models |
| 2 | [Qwen2.5-VL-32B](https://modelscale.dev/models/qwen2-5-vl-32b) | Alibaba | 26.9 | Observed 2026-11-21 · source benchlm:models |
| 3 | [Gemini 2.5 Pro](https://modelscale.dev/models/gemini-2-5-pro) | Google | 20.7 | Observed 2026-11-21 · source benchlm:models |
| 4 | [Gemini 2.5 Flash](https://modelscale.dev/models/gemini-2-5-flash) | Google | 18.7 | Observed 2026-11-21 · source benchlm:models |
| 5 | [Llama 4 Maverick](https://modelscale.dev/models/llama-4-maverick) | Meta | 13.0 | Observed 2026-11-21 · source benchlm:models |
| 6 | [Llama 4 Scout](https://modelscale.dev/models/llama-4-scout) | Meta | 8.9 | Observed 2026-11-21 · source benchlm:models |
| 7 | [Gemma 3 27B](https://modelscale.dev/models/gemma-3-27b) | Google | 0.4 | Observed 2026-11-21 · source benchlm:models |
| 7 | [Phi-4 Multimodal Instruct](https://modelscale.dev/models/phi-4-multimodal-instruct) | Microsoft | 0.4 | Observed 2026-11-21 · source benchlm:models |

Rows are ordered by the value the source published, highest first. The source does not state whether a higher value is the better result, so this document does not either: for a benchmark that measures a rate of failure, read the table from the bottom.
