Loading page…
Preferred languages
More languages supported by Google Translate
Open the selector to load the current language catalog.
Translations are provided by Google Translate.
Loading page…
Preferred languages
More languages supported by Google Translate
Open the selector to load the current language catalog.
Translations are provided by Google Translate.
Multimodal & Grounded benchmark
RefCOCO average. Every model the catalog carries a published RefCOCO (avg) value for, ranked by that value.
A referring-expression grounding benchmark averaged across RefCOCO variants to test whether a model can localize described objects correctly. 10 models in the ModelChoice catalog have a published value.
Each point is a model with a published RefCOCO (avg) value and complete token prices, on the 75% input / 25% output blend. Frontier models are spaced evenly along the cost axis.
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 |
|---|---|---|---|
| 1 | Alibaba | 92.5 | |
| 2 | Alibaba | 92.3 | |
| 3 | Alibaba | 92.0 |
What the benchmark measures and how its value is published, from the source’s own record.
RefCOCO (avg) (RefCOCO average) is a multimodal & grounded benchmark published by RefCOCO referring expression datasets. A referring-expression grounding benchmark averaged across RefCOCO variants to test whether a model can localize described objects correctly. Its evaluation set is described by the source as referring-expression grounding. The source rates its difficulty as fine-grained visual grounding.
Each model's result is published by RefCOCO referring expression datasets as grounded visual localization. ModelChoice prints that value to 1 decimal place, exactly as published, and never converts or rescales it. This page ranks the 10 models that carry a published value, highest value first; a model the source never scored is left out rather than ranked at zero. The source does not state whether a higher value is the better result, so neither does this page.
All leaderboards · Multimodal & Grounded capability leaderboard
| 4 | LiquidAI | 90.5 |
| 4 | NVIDIA | 90.5 |
| 6 | LiquidAI | 87.9 |
| 7 | Moonshot AI | 87.8 |
| 8 | ZYZAYA1-VL-8B | Zyphra | 84.3 |
| 9 | 84.1 |
| 10 | INInterfaze Beta | Interfaze | 82.1 |
Evidence key: Observed
Rows are ordered by the value RefCOCO referring expression datasets published, highest first. The source does not state whether a higher value is the better result, so this page does not either: for a benchmark that measures a rate of failure, read the table from the bottom.