Benchmark Almanac

RefCOCO / RefCOCO+ / RefCOCOg

Qwen3.5-Omni-Plus 95% leads since March 2026
In use

Model grounds a referring expression to a bounding box in a COCO image. Scored by accuracy at IoU 0.5.

Reka's 'RefCOCO-A/B' are the testA/testB splits.

Released
July 2016
Kind
Detection and captioning
Level
Everyday
Scored
Checked automatically
Answers
Short answer
Tasks
All public
Strongest model tested
Qwen3.5 397B-A17B #77 on the capability index
Last in a launch post
August 2026
Last model added
September 2026

In the authors’ words

Humans refer to objects in their environments all the time, especially in dialogue with other people. We explore generating and comprehending natural language referring expressions for objects in images. In particular, we focus on incorporating better measures of visual context into referring expression models and find that visual comparison to other objects within an image helps improve performance significantly. We also develop methods to tie the language generation process together, so that we generate expressions for all objects of a particular category jointly. Evaluation on three recent datasets - RefCOCO, RefCOCO+, and RefCOCOg, shows the advantages of our methods for both referring expression generation and comprehension.

Abstract of Modeling Context in Referring Expressions (arXiv:1608.00272, CC0)
29
Models scored in lab cards
24
Launch posts list it, from 10 labs
17
System cards report it, from 9 labs

Qwen3.5-Omni-Plus leads with 95% among models released by September 2026

Best score per model as labs report them in their own cards. Labs pick their own settings, so these compare less cleanly than independent runs.

The newest model with a score here came out in September 2026, and nobody has reported a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

29 models. Scroll sideways for the rest.

  1. 95.0 Qwen3.5-Omni-Plus
  2. 93.9 LongCat-Flash-Omni
  3. 92.5 Qwen3.6 27B
  4. 92.4 InternVL3.5-241B-A28B
  5. 92.4 Qwen3-VL-235B-A22B
  6. 92.3 Qwen3.5 397B-A17B
  7. 92.0 Qwen 3.6 35B-A3B
  8. 91.9 Intern-S1-Pro
  9. 91.3 Seed1.5-VL
  10. 91.3 GLM-4.5V
  11. 91.3 Qwen3.5-122B-A10B
  12. 90.9 Qwen3.5-27B
  13. 90.5 Nemotron 3 Nano Omni 30B-A3B
  14. 89.9 Reka Edge
  15. 89.7 Qwen3.5-9B
  16. 89.2 Qwen3.5-35B-A3B
  17. 88.6 GLM-4.6V
  18. 88.1 Qwen3.5-4B
  19. 88.1 Keye-VL-2.0-30B-A3B
  20. 87.9 LFM2.5-VL-3B
  21. 87.7 Qwen2.5-Omni-7B
  22. 87.4 GLM-4.1V-9B-Thinking
  23. 86.3 Nova 2 Omni Preview
  24. 86.0 DeepSeek-V4.1-Flash Base
  25. 84.8 Qwen3.5-2B
  26. 84.3 Cosmos3-Nano
  27. 80.1 Cosmos3-Edge
  28. 79.3 Qwen3.5-0.8B
  29. 73.2 North Micro Vision Instruct
  1. 1 Qwen3.5-Omni-Plus 95%
  2. 2 LongCat-Flash-Omni 93.9%
  3. 3 Qwen3.6 27B 92.5%
  4. 4 InternVL3.5-241B-A28B 92.4%
  5. 5 Qwen3-VL-235B-A22B 92.4%
  6. 6 Qwen3.5 397B-A17B 92.3%
  7. 7 Qwen 3.6 35B-A3B 92%
  8. 8 Intern-S1-Pro 91.9%
  9. 9 Seed1.5-VL 91.3%
  10. 10 GLM-4.5V 91.3%
  11. 11 Qwen3.5-122B-A10B 91.3%
  12. 12 Qwen3.5-27B 90.9%
  13. 13 Nemotron 3 Nano Omni 30B-A3B 90.5%
  14. 14 Reka Edge 89.9%
  15. 15 Qwen3.5-9B 89.7%
  16. 16 Qwen3.5-35B-A3B 89.2%
  17. 17 GLM-4.6V 88.6%
  18. 18 Qwen3.5-4B 88.1%
  19. 19 Keye-VL-2.0-30B-A3B 88.1%
  20. 20 LFM2.5-VL-3B 87.9%
  21. 21 Qwen2.5-Omni-7B 87.7%
  22. 22 GLM-4.1V-9B-Thinking 87.4%
  23. 23 Nova 2 Omni Preview 86.3%
  24. 24 DeepSeek-V4.1-Flash Base 86%
  25. 25 Qwen3.5-2B 84.8%
  26. 26 Cosmos3-Nano 84.3%
  27. 27 Cosmos3-Edge 80.1%
  28. 28 Qwen3.5-0.8B 79.3%
  29. 29 North Micro Vision Instruct 73.2%

Source: Labs’ model cards

#ModelScoreSettingDateSource
1 Qwen3.5-Omni-Plus 95% Mar 2026 Lab model card
2 LongCat-Flash-Omni 93.9% Oct 2025 Lab model card
3 Qwen3.6 27B 92.5% Apr 2026 Lab model card
4 InternVL3.5-241B-A28B 92.4% Aug 2025 Lab model card
5 Qwen3-VL-235B-A22B 92.4% Sep 2025 Lab model card
6 Qwen3.5 397B-A17B 92.3% Feb 2026 Lab model card
7 Qwen 3.6 35B-A3B 92% Apr 2026 Lab model card
8 Intern-S1-Pro 91.9% Feb 2026 Lab model card
9 Seed1.5-VL 91.3% May 2025 Lab model card
10 GLM-4.5V 91.3% Aug 2025 Lab model card
11 Qwen3.5-122B-A10B 91.3% Feb 2026 Lab model card
12 Qwen3.5-27B 90.9% Feb 2026 Lab model card
13 Nemotron 3 Nano Omni 30B-A3B 90.5% Apr 2026 Lab model card
14 Reka Edge 89.9% Mar 2026 Lab model card
15 Qwen3.5-9B 89.7% Mar 2026 Lab model card
16 Qwen3.5-35B-A3B 89.2% Feb 2026 Lab model card
17 GLM-4.6V 88.6% Dec 2025 Lab model card
18 Qwen3.5-4B 88.1% Mar 2026 Lab model card
19 Keye-VL-2.0-30B-A3B 88.1% May 2026 Lab model card
20 LFM2.5-VL-3B 87.9% Aug 2026 Lab model card
21 Qwen2.5-Omni-7B 87.7% Mar 2025 Lab model card
22 GLM-4.1V-9B-Thinking 87.4% Jul 2025 Lab model card
23 Nova 2 Omni Preview 86.3% Dec 2025 Lab model card
24 DeepSeek-V4.1-Flash Base 86% Sep 2026 Lab model card
25 Qwen3.5-2B 84.8% Mar 2026 Lab model card
26 Cosmos3-Nano 84.3% May 2026 Lab model card
27 Cosmos3-Edge 80.1% May 2026 Lab model card
28 Qwen3.5-0.8B 79.3% Mar 2026 Lab model card
29 North Micro Vision Instruct 73.2% Aug 2026 Lab model card

5 models have held number one in labs’ own numbers since 2023

Lab claims count from the day the card came out, independent runs from the model’s release.

Qwen ByteDance Seed Shanghai AI Lab Meituan
Lab claims Passed 85% in Mar 2025, by Qwen2.5-Omni-7B
Qwen3.5-Omni-Plus 95%

Source: Labs’ model cards

Qwen’s own number for Qwen2.5-Omni-7B passed 85% in March 2025

All the scores we have, one dot each, so a model tested at several settings appears more than once.

Reported by the lab in its model card Best so far
0% 20% 40% 60% 80% 100% 2025 2026 Near the ceiling (85%) Qwen3.5-Omni-Plus LongCat-Flash-Omni Seed1.5-VL 0% 20% 40% 60% 80% 100% 2025 2026 Near the ceiling (85%) Qwen3.5-Omni-Plus Qwen2.5-Omni-7B

Source: Labs’ model cards

2 labs put it in a launch post in the last six months

Launch posts from every lab that has listed it, one tick per post. Tall colored ticks include it, and short gray ones leave it out.

Qwen
ByteDance Seed
Z.ai
Meituan
Amazon
Shanghai AI Lab
Reka
Kuaishou
Cohere
Liquid AI
DateLaunch postScoreAs the post writes it
Aug 12, 2026 LFM2.5-VL-3B 87.9 RefCOCO-avg
Aug 12, 2026 North Micro Vision Instruct 73.2 RefCOCO avg (average of 8 RefCOCO, RefCOCO+ and RefCOCOg splits)
May 25, 2026 Keye-VL-2.0-30B-A3B 88.1 RefCOCO (avg)
Apr 22, 2026 Qwen3.6 27B 92.5 RefCOCO avg
Apr 15, 2026 Qwen 3.6 35B-A3B 92 RefCOCO(avg)
Mar 30, 2026 Qwen3.5-Omni-Plus 95 RefCOCO(avg)
Mar 11, 2026 Reka Edge 89.9 Mean of RefCOCO-A 93.13 / RefCOCO-B 86.70
Mar 2, 2026 Qwen3.5-0.8B 79.3 RefCOCO(avg) 79.3/77.8 (with CI / without CI)
Mar 2, 2026 Qwen3.5-2B 84.8 RefCOCO(avg) 84.8/84.3 (with CI / without CI)
Mar 2, 2026 Qwen3.5-4B 88.1 RefCOCO(avg)
Mar 2, 2026 Qwen3.5-9B 89.7 RefCOCO(avg)
Feb 24, 2026 Qwen3.5-35B-A3B 89.2 RefCOCO(avg)
Feb 24, 2026 Qwen3.5-27B 90.9 RefCOCO(avg)
Feb 24, 2026 Qwen3.5-122B-A10B 91.3 RefCOCO(avg)
Feb 16, 2026 Qwen3.5 397B-A17B 92.3 RefCOCO(avg)
Feb 2, 2026 Intern-S1-Pro 91.9 RefCOCO-avg
Dec 8, 2025 GLM-4.6V 88.6 RefCOCO-avg (val)
Dec 2, 2025 Nova 2 Omni Preview 86.3 RefCOCOg acc@0.5
Oct 31, 2025 LongCat-Flash-Omni 93.9 RefCOCO-avg
Sep 23, 2025 Qwen3-VL-235B-A22B 92.4 RefCOCO-avg
Aug 11, 2025 GLM-4.5V 91.3 RefCOCO-avg (val)
Jul 1, 2025 GLM-4.1V-9B-Thinking 87.4 RefCOCO-avg (val)
May 12, 2025 Seed1.5-VL 91.3 RefCOCO-avg
Mar 27, 2025 Qwen2.5-Omni-7B 87.7 Mean of Refcoco val/testA/testB, Refcoco+ val/testA/testB, Refcocog val/test (90.5/93.5/86.6/85.4/91.0/79.3/87.4/87.9)

17 system cards report it, from 9 labs

Besides launch posts, we read the system cards, model cards and technical reports labs publish with a release. They report far more, including safety and frontier-risk tests.

DateDocumentScoreAs the document writes itSection
Sep 10, 2026 DeepSeek-V4.1-Flash Base model card 86 RefCOCO-avg, Acc@0.5, 0-shot Capabilities
Jun 9, 2026 Keye-VL-2.0-30B-A3B technical report 88.1 RefCOCO (avg) (Figure 5, printed value) Capabilities
May 31, 2026 Cosmos3-Nano technical report 84.3 RefCOCO, Cosmos3-Nano 8B column, Table 10 Capabilities
May 31, 2026 Cosmos3-Edge technical report 80.1 RefCOCO, Cosmos3-Edge 2B column, Table 10 Capabilities
Apr 27, 2026 Nemotron 3 Nano Omni 30B-A3B technical report 90.5 RefCOCO, reasoning on, 80.6 reasoning off (Table 7) Capabilities
Apr 17, 2026 Qwen3.5-Omni-Plus technical report 95 RefCOCO(avg) (Table 6) Capabilities
Mar 26, 2026 Intern-S1-Pro technical report 91.9 RefCOCO-avg, non-thinking (Table 2) Capabilities
Jan 1, 2026 GLM-4.6V technical report 88.6 RefCOCO-avg (val), GLM-4.6V thinking (Table 2) Capabilities
Dec 2, 2025 Nova 2 Omni Preview technical report 86.3 RefCOCOg acc@0.5 (Table 6) Capabilities
Nov 26, 2025 Qwen3-VL-235B-A22B technical report 92.1 RefCOCO-avg (Table 2) Capabilities
Oct 31, 2025 LongCat-Flash-Omni technical report 93.9 RefCOCO-avg, average of RefCOCO and RefCOCO+ (Table 5) Capabilities
Aug 25, 2025 InternVL3.5-241B-A28B technical report 92.4 RefCOCO, RefCOCO+ and RefCOCOg Overall (Table 7). RefCOCO val 94.1, testA 96.3, testB 91.5. RefCOCO+ val 91.6, testA 94.6, testB 86.9. RefCOCOg val 92.0, test 92.1 Capabilities
Aug 13, 2025 GLM-4.5V technical report 91.3 RefCOCO-avg (val), GLM-4.5V thinking (Table 2) Capabilities
Jul 1, 2025 GLM-4.1V-9B-Thinking technical report 85.3 RefCOCO-avg (val), GLM-4.1V-9B-Thinking thinking (Table 2). Table 3 prints 87.4 Capabilities
May 11, 2025 Seed1.5-VL technical report 91.3 RefCOCO-avg, thinking (Table 6). Non-thinking 91.6 Capabilities
Mar 26, 2025 Qwen2.5-Omni-7B technical report Refcoco val 90.5, testA 93.5, testB 86.6. Refcoco+ val 85.4, testA 91.0, testB 79.3. Refcocog val 87.4, test 87.9. No overall score (Table 6) Capabilities
Feb 19, 2025 Qwen2.5-VL-72B technical report RefCOCO val 92.7, testA 94.6, testB 89.7, RefCOCO+ val 88.9, testA 92.2, testB 83.7, RefCOCOg val 89.9, test 90.3, no overall (Table 6) Capabilities

More in Vision

All 281 in Vision
  • MMMU In use 11.5K college-level multimodal questions across 30 subjects. Multiple-choice and open accuracy. GPT-5.1 85.4%
  • MMMU-Pro In use Harder MMMU: college-level multimodal questions with more options and vision-only inputs to remove text shortcuts. GPT-5.6 Sol 84.6%
  • CharXiv In use Chart understanding on real arXiv figures. Reasoning questions require synthesizing information from complex charts. Claude Fable 5 93.5%
  • MATH-Vision (MathVision) In use 3,040 competition math problems with visual context. Answer accuracy. Qwen 3.8 Max 97.7%
  • MMStar In use Models answer 1,500 human-curated image questions that cannot be solved without the image, across 6 core capabilities. Scored by accuracy. Qwen 3.8 Max 85.9%
  • ChartQA In use Models answer human-written and generated questions about chart images that need visual and arithmetic reasoning. Scored by relaxed accuracy with a 5 percent numeric tolerance. Command A Vision 90.9%