# RefCOCO (avg) leaderboard

Multimodal & Grounded 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/multimodal-grounded-refcoco-avg

RefCOCO average. Every model the catalog carries a published RefCOCO (avg) value for, ranked by that value.

## Benchmark record

| Field | Value |
| --- | --- |
| Category | Multimodal & Grounded |
| Measure | Grounded visual localization |
| Tasks | Referring-expression grounding |
| Difficulty | Fine-grained visual grounding |
| Published by | [RefCOCO referring expression datasets](https://github.com/lichengunc/refer) |

A referring-expression grounding benchmark averaged across RefCOCO variants to test whether a model can localize described objects correctly.

## RefCOCO (avg) ranking

10 models with a published RefCOCO (avg) 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 | Grounded visual localization | Evidence |
| --- | --- | --- | --- | --- |
| 1 | [Qwen3.6-27B](https://modelscale.dev/models/qwen3-6-27b) | Alibaba | 92.5 | Observed 2026-11-21 · source benchlm:models |
| 2 | [Qwen3.5 397B](https://modelscale.dev/models/qwen3-5-397b) | Alibaba | 92.3 | Observed 2026-11-21 · source benchlm:models |
| 3 | [Qwen3.6-35B-A3B](https://modelscale.dev/models/qwen3-6-35b-a3b) | Alibaba | 92.0 | Observed 2026-11-21 · source benchlm:models |
| 4 | [LFM3-VL-2B](https://modelscale.dev/models/lfm3-vl-2b) | LiquidAI | 90.5 | Observed 2026-11-21 · source benchlm:models |
| 4 | [Nemotron 3 Nano Omni 30B A3B](https://modelscale.dev/models/nemotron-3-nano-omni-30b-a3b) | NVIDIA | 90.5 | Observed 2026-11-21 · source benchlm:models |
| 6 | [LFM2.5-VL-3B](https://modelscale.dev/models/lfm2-5-vl-3b) | LiquidAI | 87.9 | Observed 2026-11-21 · source benchlm:models |
| 7 | [Kimi K2.5](https://modelscale.dev/models/kimi-k2-5) | Moonshot AI | 87.8 | Observed 2026-11-21 · source benchlm:models |
| 8 | [ZAYA1-VL-8B](https://modelscale.dev/models/zaya1-vl-8b) | Zyphra | 84.3 | Observed 2026-11-21 · source benchlm:models |
| 9 | [Gemini 3 Pro](https://modelscale.dev/models/gemini-3-pro) | Google | 84.1 | Observed 2026-11-21 · source benchlm:models |
| 10 | [Interfaze Beta](https://modelscale.dev/models/interfaze-beta) | Interfaze | 82.1 | 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.
