OBI-Bench
Models locate inscriptions, rejoin bone fragments, classify characters, retrieve matches, and explain glyphs. The authors publish separate scores for each task.
The authors publish task results for 23 vision language models across the suite. The benchmark draws 5,523 images from 11 sources.
- Made by
- Shanghai Jiao Tong University
- Released
- October 2024
- Kind
- Documents, charts and text in images
- Level
- Expert
- Scored
- Checked automatically
- Answers
- Open-ended
- Tasks
- All public
- Measures
- Capability
- Language
- Chinese
- Interaction
- Single turn
- Items from
- Real work and use, Another benchmark
- Refresh
- One fixed set
- Run by
- Anyone
- In views
- Vision, Image, video and 3D generation
- Strongest model tested
- GPT-4o Aug 2024 #167 on the capability index
- Last model added
- August 2024
- Scores refreshed
- November 20th, 2026
GPT-4o leads with 0.61 among models released by August 2024 GPT-4o leads with 0.61 among models released by August 2024
Best score per entry from the benchmark's paper, read on November 20th, 2026. A paper’s table doesn’t add newer models. The paper gives no overall score, so the headline is cosine similarity on the O2BR recognition question “What”. Where a model ran with several settings or harnesses, we keep its best. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders. This view shows only models we track, each at its best.
The newest model with a score here came out in August 2024, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:
- GPT-6.2 Astra
- Claude Fable 5.2
-
Gemini 4 Carbon
- GPT-6 Astra
- and 16 more
- 0.61 GPT-4o
- 0.61 Qwen-VL-Max (ver. 0809)
- 0.58
InternVL2-Llama3-76B
- 0.57 Gemini 1.5 Pro (snapshot unspecified)
- 0.57
InternVL2-40B
- 0.56 MiniCPM-V-2_6
- 0.54 GPT-4V
- 0.54 GLM-4V-9B
- 0.53 CogVLM2-19B-Chat
- 0.53 GLM-4V
- 0.52 xGen-MM (Instruct-interleave-4B)
- 0.52
InternVL2-8B
- 0.51 DeepSeek-VL (DeepSeek-LLM-7B)
- 0.51
InternLM-XComposer2-VL-7B
- 0.48
LLaVA-NeXT-72B
- 0.48 Moondream2 (ver. 0728)
- 0.48 llama3-llava-next-8b
- 0.45 Qwen-VL (Qwen-7B)
- 0.44
LLaVA-v1.5-13B
- 0.42 llava-v1.5-7b
- 0.41 Gemini 1.5 Flash (snapshot unspecified)
- 0.33
mPLUG-Owl3-7B
- 0.32
Idefics2-8B
- 1 GPT-4o Aug 2024 0.61
- 2 Qwen-VL-Max (ver. 0809) 0.61
- 3
InternVL2-Llama3-76B 0.58
- 4 Gemini 1.5 Pro (snapshot unspecified) 0.57
- 5
InternVL2-40B 0.57
- 6 MiniCPM-V-2_6 0.56
- 7 GPT-4V 0.54
- 8 GLM-4V-9B 0.54
- 9 CogVLM2-19B-Chat 0.53
- 10 GLM-4V 0.53
- 11 xGen-MM (Instruct-interleave-4B) 0.52
- 12
InternVL2-8B 0.52
- 13 DeepSeek-VL (DeepSeek-LLM-7B) 0.51
- 14
InternLM-XComposer2-VL-7B 0.51
- 15
LLaVA-NeXT-72B 0.48
- 16 Moondream2 (ver. 0728) 0.48
- 17 llama3-llava-next-8b 0.48
- 18 Qwen-VL (Qwen-7B) 0.45
- 19
LLaVA-v1.5-13B 0.44
- 20 llava-v1.5-7b 0.42
- 21 Gemini 1.5 Flash (snapshot unspecified) 0.41
- 22
mPLUG-Owl3-7B 0.33
- 23
Idefics2-8B 0.32
Source: The OBI-Bench paper
- 1 GPT-4o Aug 2024 0.61
- 2
InternVL2-Llama3-76B 0.58
- 3
InternVL2-40B 0.57
- 4 MiniCPM-V-2_6 0.56
- 5 GPT-4V 0.54
- 6 GLM-4V-9B 0.54
- 7 CogVLM2-19B-Chat 0.53
- 8
InternVL2-8B 0.52
- 9
InternLM-XComposer2-VL-7B 0.51
- 10
LLaVA-NeXT-72B 0.48
- 11 llama3-llava-next-8b 0.48
- 12
LLaVA-v1.5-13B 0.44
- 13 llava-v1.5-7b 0.42
- 14
mPLUG-Owl3-7B 0.33
- 15
Idefics2-8B 0.32
Source: The OBI-Bench paper
| # | Entry | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | GPT-4o Aug 2024 | 0.61 | Aug 2024 | Official leaderboard | |
| 2 | Qwen-VL-Max (ver. 0809) | 0.61 | Official leaderboard | ||
| 3 | | 0.58 | Jul 2024 | Official leaderboard | |
| 4 | Gemini 1.5 Pro (snapshot unspecified) | 0.57 | Official leaderboard | ||
| 5 | | 0.57 | Jul 2024 | Official leaderboard | |
| 6 | MiniCPM-V-2_6 | 0.56 | Aug 2024 | Official leaderboard | |
| 7 | GPT-4V | 0.54 | Nov 2023 | Official leaderboard | |
| 8 | GLM-4V-9B | 0.54 | Jun 2024 | Official leaderboard | |
| 9 | CogVLM2-19B-Chat | 0.53 | May 2024 | Official leaderboard | |
| 10 | GLM-4V | 0.53 | Official leaderboard | ||
| 11 | xGen-MM (Instruct-interleave-4B) | 0.52 | Official leaderboard | ||
| 12 | | 0.52 | Jul 2024 | Official leaderboard | |
| 13 | DeepSeek-VL (DeepSeek-LLM-7B) | 0.51 | Official leaderboard | ||
| 14 | | 0.51 | Jan 2024 | Official leaderboard | |
| 15 | | 0.48 | May 2024 | Official leaderboard | |
| 16 | Moondream2 (ver. 0728) | 0.48 | Official leaderboard | ||
| 17 | llama3-llava-next-8b | 0.48 | Apr 2024 | Official leaderboard | |
| 18 | Qwen-VL (Qwen-7B) | 0.45 | Official leaderboard | ||
| 19 | | 0.44 | Oct 2023 | Official leaderboard | |
| 20 | llava-v1.5-7b | 0.42 | Oct 2023 | Official leaderboard | |
| 21 | Gemini 1.5 Flash (snapshot unspecified) | 0.41 | Official leaderboard | ||
| 22 | | 0.33 | Aug 2024 | Official leaderboard | |
| 23 | | 0.32 | Apr 2024 | Official leaderboard |
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | GPT-4o Aug 2024 | 0.61 | Aug 2024 | Official leaderboard | |
| 2 | | 0.58 | Jul 2024 | Official leaderboard | |
| 3 | | 0.57 | Jul 2024 | Official leaderboard | |
| 4 | MiniCPM-V-2_6 | 0.56 | Aug 2024 | Official leaderboard | |
| 5 | GPT-4V | 0.54 | Nov 2023 | Official leaderboard | |
| 6 | GLM-4V-9B | 0.54 | Jun 2024 | Official leaderboard | |
| 7 | CogVLM2-19B-Chat | 0.53 | May 2024 | Official leaderboard | |
| 8 | | 0.52 | Jul 2024 | Official leaderboard | |
| 9 | | 0.51 | Jan 2024 | Official leaderboard | |
| 10 | | 0.48 | May 2024 | Official leaderboard | |
| 11 | llama3-llava-next-8b | 0.48 | Apr 2024 | Official leaderboard | |
| 12 | | 0.44 | Oct 2023 | Official leaderboard | |
| 13 | llava-v1.5-7b | 0.42 | Oct 2023 | Official leaderboard | |
| 14 | | 0.33 | Aug 2024 | Official leaderboard | |
| 15 | | 0.32 | Apr 2024 | Official leaderboard |
Rankings by part
We read 38 parts from the benchmark’s paper in 5 groups: recognition, rejoining, classification, retrieval and deciphering. Each ranks models on one part of the test. GPT-4o leads 27 of them, and 5 other models lead the rest. Arrows show how far a model moved from its overall rank among the models we track.
O2BR: What
15 models we track have a score here.
- 1 GPT-4o Aug 2024 0.61
- 2
InternVL2-Llama3-76B 0.58
- 3
InternVL2-40B 0.57
- 4 MiniCPM-V-2_6 0.56
- 5 GPT-4V 0.54
- 6 GLM-4V-9B 0.54
- 7 CogVLM2-19B-Chat 0.53
- 8
InternVL2-8B 0.52
- 9
InternLM-XComposer2-VL-7B 0.51
- 10
LLaVA-NeXT-72B 0.48
O2BR: Yes-or-No
15 models we track have a score here.
- 1 GPT-4o Aug 2024 99.95%
- 2
InternLM-XComposer2-VL-7B ↑7 99.9%
- 3
mPLUG-Owl3-7B ↑11 99.9%
- 4
LLaVA-NeXT-72B ↑6 99.8%
- 5
InternVL2-Llama3-76B ↓3 99.7%
- 6 GPT-4V ↓1 99.3%
- 7
LLaVA-v1.5-13B ↑5 99%
- 8
InternVL2-40B ↓5 98.3%
- 9
Idefics2-8B ↑6 95.9%
- 10
InternVL2-8B ↓2 95%
O2BR: How
15 models we track have a score here.
- 1 GPT-4o Aug 2024 0.40
- 2 GPT-4V ↑3 0.42
- 3
InternVL2-Llama3-76B ↓1 0.43
- 4
mPLUG-Owl3-7B ↑10 0.45
- 5
InternVL2-40B ↓2 0.46
- 6
InternVL2-8B ↑2 0.46
- 7 GLM-4V-9B ↓1 0.48
- 8 MiniCPM-V-2_6 ↓4 0.48
- 9
Idefics2-8B ↑6 0.49
- 10 llama3-llava-next-8b ↑1 0.55
O2BR: Where
12 models we track have a score here.
- 1
InternVL2-Llama3-76B ↑1 0.10
- 2
mPLUG-Owl3-7B ↑12 0.08
- 3
InternVL2-40B 0.08
- 4
LLaVA-v1.5-13B ↑8 0.08
- 5 llava-v1.5-7b ↑8 0.07
- 6
LLaVA-NeXT-72B ↑4 0.04
- 7 MiniCPM-V-2_6 ↓3 0.04
- 8
InternLM-XComposer2-VL-7B ↑1 0.00
- 9 GPT-4o Aug 2024 ↓8 0.00
- 10 GPT-4V ↓5 0.00
YinQiWenYuan detection: What
15 models we track have a score here.
- 1 CogVLM2-19B-Chat ↑6 0.40
- 2
InternVL2-Llama3-76B 0.39
- 3 MiniCPM-V-2_6 ↑1 0.38
- 4 GPT-4o Aug 2024 ↓3 0.37
- 5 GPT-4V 0.37
- 6
InternVL2-40B ↓3 0.36
- 7
InternVL2-8B ↑1 0.34
- 8
LLaVA-NeXT-72B ↑2 0.33
- 9 llama3-llava-next-8b ↑2 0.31
- 10
LLaVA-v1.5-13B ↑2 0.29
YinQiWenYuan detection: Yes-or-No
15 models we track have a score here.
- 1
InternLM-XComposer2-VL-7B ↑8 99.95%
- 2
mPLUG-Owl3-7B ↑12 99.95%
- 3 GPT-4o Aug 2024 ↓2 99.9%
- 4
LLaVA-NeXT-72B ↑6 99.8%
- 5
InternVL2-Llama3-76B ↓3 99.8%
- 6 GPT-4V ↓1 99.7%
- 7 MiniCPM-V-2_6 ↓3 99.1%
- 8
InternVL2-40B ↓5 99%
- 9
LLaVA-v1.5-13B ↑3 98%
- 10
InternVL2-8B ↓2 98%
YinQiWenYuan detection: How
15 models we track have a score here.
- 1 GPT-4o Aug 2024 0.35
- 2 GLM-4V-9B ↑4 0.39
- 3 llama3-llava-next-8b ↑8 0.43
- 4 GPT-4V ↑1 0.44
- 5
Idefics2-8B ↑10 0.51
- 6
InternVL2-Llama3-76B ↓4 0.53
- 7 CogVLM2-19B-Chat 0.56
- 8
mPLUG-Owl3-7B ↑6 0.66
- 9
InternVL2-40B ↓6 0.67
- 10
LLaVA-NeXT-72B 1.06
YinQiWenYuan detection: Where
13 models we track have a score here.
- 1
InternVL2-Llama3-76B ↑1 0.06
- 2
LLaVA-NeXT-72B ↑8 0.06
- 3
mPLUG-Owl3-7B ↑11 0.05
- 4
LLaVA-v1.5-13B ↑8 0.05
- 5
InternVL2-40B ↓2 0.05
- 6 llava-v1.5-7b ↑7 0.05
- 7
InternLM-XComposer2-VL-7B ↑2 0.03
- 8 llama3-llava-next-8b ↑3 0.02
- 9 GPT-4o Aug 2024 ↓8 0.02
- 10 GPT-4V ↓5 0.02
OBI-rejoin: Yes-or-No
10 models we track have a score here.
- 1 GPT-4o Aug 2024 32.2%
- 2 GPT-4V ↑3 27.6%
- 3
InternVL2-Llama3-76B ↓1 20.1%
- 4
LLaVA-NeXT-72B ↑6 17.4%
- 5
InternVL2-40B ↓2 16%
- 6
InternVL2-8B ↑2 14.7%
- 7
mPLUG-Owl3-7B ↑7 14.5%
- 8 MiniCPM-V-2_6 ↓4 13.2%
- 9 llama3-llava-next-8b ↑2 11.4%
- 10
Idefics2-8B ↑5 7.38%
OBI-rejoin: Acc@1
10 models we track have a score here.
- 1 GPT-4o Aug 2024 29.1%
- 2 GPT-4V ↑3 26.9%
- 3
InternVL2-Llama3-76B ↓1 5.18%
- 4
LLaVA-NeXT-72B ↑6 4.97%
- 5 llama3-llava-next-8b ↑6 3.93%
- 6
InternVL2-8B ↑2 3.73%
- 7
Idefics2-8B ↑8 3.52%
- 8
mPLUG-Owl3-7B ↑6 3.52%
- 9
InternVL2-40B ↓6 3.31%
- 10 MiniCPM-V-2_6 ↓6 2.69%
OBI-rejoin: Acc@5
10 models we track have a score here.
- 1 GPT-4o Aug 2024 48.4%
- 2 GPT-4V ↑3 43.8%
- 3
InternVL2-Llama3-76B ↓1 17%
- 4
LLaVA-NeXT-72B ↑6 14.3%
- 5
mPLUG-Owl3-7B ↑9 12.6%
- 6
Idefics2-8B ↑9 12.4%
- 7 MiniCPM-V-2_6 ↓3 11.2%
- 8
InternVL2-40B ↓5 11%
- 9
InternVL2-8B ↓1 10.1%
- 10 llama3-llava-next-8b ↑1 9.52%
OBI-rejoin: Acc@10
10 models we track have a score here.
- 1 GPT-4o Aug 2024 78.5%
- 2 GPT-4V ↑3 73.8%
- 3
InternVL2-Llama3-76B ↓1 31.7%
- 4
LLaVA-NeXT-72B ↑6 27.7%
- 5
InternVL2-40B ↓2 24.8%
- 6
InternVL2-8B ↑2 17.8%
- 7
mPLUG-Owl3-7B ↑7 17.6%
- 8 MiniCPM-V-2_6 ↓4 17%
- 9 llama3-llava-next-8b ↑2 16.8%
- 10
Idefics2-8B ↑5 14.9%
HWOBC: Yes-or-No
10 models we track have a score here.
- 1 GPT-4o Aug 2024 72.8%
- 2 GPT-4V ↑3 69.5%
- 3
LLaVA-NeXT-72B ↑7 46.3%
- 4
Idefics2-8B ↑11 44.8%
- 5
InternVL2-Llama3-76B ↓3 44.8%
- 6 llama3-llava-next-8b ↑5 44%
- 7
InternVL2-40B ↓4 42.8%
- 8
InternVL2-8B 42.3%
- 9 MiniCPM-V-2_6 ↓5 40.8%
- 10
mPLUG-Owl3-7B ↑4 39.3%
HWOBC: Acc@1
10 models we track have a score here.
- 1 GPT-4o Aug 2024 89.8%
- 2 GPT-4V ↑3 86.8%
- 3
InternVL2-Llama3-76B ↓1 53.8%
- 4
LLaVA-NeXT-72B ↑6 51.8%
- 5
InternVL2-40B ↓2 49.8%
- 6
InternVL2-8B ↑2 47.8%
- 7
Idefics2-8B ↑8 46.8%
- 8 llama3-llava-next-8b ↑3 46.8%
- 9 MiniCPM-V-2_6 ↓5 45.8%
- 10
mPLUG-Owl3-7B ↑4 44.8%
HWOBC: Acc@5
10 models we track have a score here.
- 1 GPT-4V ↑4 100%
- 2 GPT-4o Aug 2024 ↓1 100%
- 3
InternVL2-Llama3-76B ↓1 69.8%
- 4
LLaVA-NeXT-72B ↑6 66.8%
- 5
InternVL2-40B ↓2 63.8%
- 6
InternVL2-8B ↑2 59.8%
- 7
Idefics2-8B ↑8 58.8%
- 8 MiniCPM-V-2_6 ↓4 56.8%
- 9
mPLUG-Owl3-7B ↑5 56.3%
- 10 llama3-llava-next-8b ↑1 53.8%
Oracle-50k: Yes-or-No
10 models we track have a score here.
- 1 GPT-4o Aug 2024 74.5%
- 2 GPT-4V ↑3 66%
- 3
InternVL2-Llama3-76B ↓1 47.5%
- 4
LLaVA-NeXT-72B ↑6 45.5%
- 5
Idefics2-8B ↑10 44.5%
- 6
InternVL2-40B ↓3 43.5%
- 7 MiniCPM-V-2_6 ↓3 42.5%
- 8 llama3-llava-next-8b ↑3 42%
- 9
InternVL2-8B ↓1 41.8%
- 10
mPLUG-Owl3-7B ↑4 39.8%
Oracle-50k: Acc@1
10 models we track have a score here.
- 1 GPT-4o Aug 2024 90.3%
- 2 GPT-4V ↑3 88.3%
- 3
InternVL2-Llama3-76B ↓1 55%
- 4
LLaVA-NeXT-72B ↑6 52%
- 5
InternVL2-40B ↓2 50%
- 6
InternVL2-8B ↑2 49%
- 7
Idefics2-8B ↑8 46.5%
- 8 llama3-llava-next-8b ↑3 46.3%
- 9
mPLUG-Owl3-7B ↑5 46.3%
- 10 MiniCPM-V-2_6 ↓6 45%
Oracle-50k: Acc@5
10 models we track have a score here.
- 1 GPT-4V ↑4 100%
- 2 GPT-4o Aug 2024 ↓1 100%
- 3
InternVL2-Llama3-76B ↓1 69%
- 4
LLaVA-NeXT-72B ↑6 67.8%
- 5
InternVL2-40B ↓2 62.3%
- 6
InternVL2-8B ↑2 59%
- 7
Idefics2-8B ↑8 58.8%
- 8 llama3-llava-next-8b ↑3 56.8%
- 9 MiniCPM-V-2_6 ↓5 55.8%
- 10
mPLUG-Owl3-7B ↑4 52.5%
OBI125: Yes-or-No
10 models we track have a score here.
- 1 GPT-4o Aug 2024 62.5%
- 2 GPT-4V ↑3 57.8%
- 3
InternVL2-Llama3-76B ↓1 43.3%
- 4
LLaVA-NeXT-72B ↑6 41%
- 5
InternVL2-40B ↓2 40.3%
- 6
Idefics2-8B ↑9 39.8%
- 7 llama3-llava-next-8b ↑4 38.8%
- 8
InternVL2-8B 38.8%
- 9
mPLUG-Owl3-7B ↑5 38.8%
- 10 MiniCPM-V-2_6 ↓6 38.5%
OBI125: Acc@1
10 models we track have a score here.
- 1 GPT-4o Aug 2024 75.5%
- 2 GPT-4V ↑3 70.8%
- 3
InternVL2-Llama3-76B ↓1 50.8%
- 4
LLaVA-NeXT-72B ↑6 49%
- 5
InternVL2-40B ↓2 48.8%
- 6
InternVL2-8B ↑2 47.8%
- 7
Idefics2-8B ↑8 44%
- 8
mPLUG-Owl3-7B ↑6 43.8%
- 9 llama3-llava-next-8b ↑2 42.8%
- 10 MiniCPM-V-2_6 ↓6 42.8%
OBI125: Acc@5
10 models we track have a score here.
- 1 GPT-4o Aug 2024 93.5%
- 2 GPT-4V ↑3 91.8%
- 3
InternVL2-Llama3-76B ↓1 66.8%
- 4
LLaVA-NeXT-72B ↑6 64.3%
- 5
InternVL2-40B ↓2 59.8%
- 6
InternVL2-8B ↑2 56.5%
- 7 llama3-llava-next-8b ↑4 54.8%
- 8
Idefics2-8B ↑7 53.8%
- 9
mPLUG-Owl3-7B ↑5 53.5%
- 10 MiniCPM-V-2_6 ↓6 51.8%
OBC306: mAP@|Yes|
10 models we track have a score here.
- 1 GPT-4o Aug 2024 0.46
- 2 GPT-4V ↑3 0.42
- 3
InternVL2-Llama3-76B ↓1 0.36
- 4
LLaVA-NeXT-72B ↑6 0.35
- 5
InternVL2-40B ↓2 0.33
- 6 MiniCPM-V-2_6 ↓2 0.29
- 7
Idefics2-8B ↑8 0.28
- 8
InternVL2-8B 0.28
- 9
mPLUG-Owl3-7B ↑5 0.28
- 10 llama3-llava-next-8b ↑1 0.28
OBC306: Recall@1
10 models we track have a score here.
- 1 GPT-4o Aug 2024 0.23
- 2 GPT-4V ↑3 0.20
- 3
LLaVA-NeXT-72B ↑7 0.15
- 4
InternVL2-Llama3-76B ↓2 0.15
- 5
InternVL2-40B ↓2 0.13
- 6
InternVL2-8B ↑2 0.10
- 7
Idefics2-8B ↑8 0.09
- 8 llama3-llava-next-8b ↑3 0.07
- 9
mPLUG-Owl3-7B ↑5 0.07
- 10 MiniCPM-V-2_6 ↓6 0.07
OBC306: Recall@3
10 models we track have a score here.
- 1 GPT-4o Aug 2024 0.69
- 2 GPT-4V ↑3 0.65
- 3
LLaVA-NeXT-72B ↑7 0.47
- 4
InternVL2-Llama3-76B ↓2 0.46
- 5
InternVL2-40B ↓2 0.45
- 6
InternVL2-8B ↑2 0.37
- 7
mPLUG-Owl3-7B ↑7 0.37
- 8 MiniCPM-V-2_6 ↓4 0.36
- 9 llama3-llava-next-8b ↑2 0.36
- 10
Idefics2-8B ↑5 0.35
OBC306: Recall@10
10 models we track have a score here.
- 1 GPT-4V ↑4 1.00
- 2
LLaVA-NeXT-72B ↑8 1.00
- 3
InternVL2-40B 1.00
- 4
InternVL2-Llama3-76B ↓2 1.00
- 5 GPT-4o Aug 2024 ↓4 1.00
- 6 llama3-llava-next-8b ↑5 0.93
- 7
InternVL2-8B ↑1 0.93
- 8 MiniCPM-V-2_6 ↓4 0.89
- 9
Idefics2-8B ↑6 0.88
- 10
mPLUG-Owl3-7B ↑4 0.87
OBC306: mAP@5
10 models we track have a score here.
- 1 GPT-4o Aug 2024 0.69
- 2 GPT-4V ↑3 0.62
- 3
LLaVA-NeXT-72B ↑7 0.56
- 4
InternVL2-Llama3-76B ↓2 0.52
- 5
InternVL2-40B ↓2 0.48
- 6
InternVL2-8B ↑2 0.42
- 7
Idefics2-8B ↑8 0.40
- 8
mPLUG-Owl3-7B ↑6 0.38
- 9 MiniCPM-V-2_6 ↓5 0.37
- 10 llama3-llava-next-8b ↑1 0.35
OBI-IJDH: mAP@|Yes|
10 models we track have a score here.
- 1 GPT-4o Aug 2024 0.61
- 2 GPT-4V ↑3 0.57
- 3
InternVL2-Llama3-76B ↓1 0.43
- 4
LLaVA-NeXT-72B ↑6 0.41
- 5
InternVL2-40B ↓2 0.39
- 6
InternVL2-8B ↑2 0.36
- 7 llama3-llava-next-8b ↑4 0.36
- 8
Idefics2-8B ↑7 0.36
- 9 MiniCPM-V-2_6 ↓5 0.36
- 10
mPLUG-Owl3-7B ↑4 0.35
OBI-IJDH: Recall@1
10 models we track have a score here.
- 1
LLaVA-NeXT-72B ↑9 0.25
- 2
InternVL2-40B ↑1 0.25
- 3
InternVL2-Llama3-76B ↓1 0.25
- 4 GPT-4o Aug 2024 ↓3 0.25
- 5 GPT-4V 0.23
- 6 llama3-llava-next-8b ↑5 0.23
- 7
InternVL2-8B ↑1 0.23
- 8 MiniCPM-V-2_6 ↓4 0.23
- 9
mPLUG-Owl3-7B ↑5 0.23
- 10
Idefics2-8B ↑5 0.20
OBI-IJDH: Recall@3
10 models we track have a score here.
- 1 GPT-4o Aug 2024 0.71
- 2 GPT-4V ↑3 0.68
- 3
InternVL2-Llama3-76B ↓1 0.68
- 4
LLaVA-NeXT-72B ↑6 0.67
- 5
InternVL2-40B ↓2 0.66
- 6
InternVL2-8B ↑2 0.65
- 7
Idefics2-8B ↑8 0.62
- 8 llama3-llava-next-8b ↑3 0.61
- 9
mPLUG-Owl3-7B ↑5 0.59
- 10 MiniCPM-V-2_6 ↓6 0.58
OBI-IJDH: Recall@10
10 models we track have a score here.
- 1 GPT-4V ↑4 1.00
- 2
Idefics2-8B ↑13 1.00
- 3 llama3-llava-next-8b ↑8 1.00
- 4
LLaVA-NeXT-72B ↑6 1.00
- 5
InternVL2-8B ↑3 1.00
- 6
InternVL2-40B ↓3 1.00
- 7
InternVL2-Llama3-76B ↓5 1.00
- 8 GPT-4o Aug 2024 ↓7 1.00
- 9
mPLUG-Owl3-7B ↑5 0.95
- 10 MiniCPM-V-2_6 ↓6 0.93
OBI-IJDH: mAP@5
10 models we track have a score here.
- 1 GPT-4o Aug 2024 0.80
- 2 GPT-4V ↑3 0.74
- 3
InternVL2-Llama3-76B ↓1 0.72
- 4
LLaVA-NeXT-72B ↑6 0.70
- 5
InternVL2-8B ↑3 0.68
- 6
InternVL2-40B ↓3 0.68
- 7
Idefics2-8B ↑8 0.66
- 8 MiniCPM-V-2_6 ↓4 0.66
- 9
mPLUG-Owl3-7B ↑5 0.62
- 10 llama3-llava-next-8b ↑1 0.60
HUST-OBS: Tier-1
15 models we track have a score here.
- 1 GPT-4o Aug 2024 0.39
- 2 GPT-4V ↑3 0.38
- 3
InternLM-XComposer2-VL-7B ↑6 0.30
- 4
InternVL2-Llama3-76B ↓2 0.23
- 5
InternVL2-40B ↓2 0.20
- 6
InternVL2-8B ↑2 0.17
- 7
Idefics2-8B ↑8 0.16
- 8
mPLUG-Owl3-7B ↑6 0.16
- 9
LLaVA-NeXT-72B ↑1 0.15
- 10 MiniCPM-V-2_6 ↓6 0.14
HUST-OBS: Tier-2
15 models we track have a score here.
- 1 GPT-4V ↑4 0.38
- 2 GPT-4o Aug 2024 ↓1 0.35
- 3
InternLM-XComposer2-VL-7B ↑6 0.31
- 4
InternVL2-Llama3-76B ↓2 0.21
- 5
InternVL2-40B ↓2 0.19
- 6
mPLUG-Owl3-7B ↑8 0.18
- 7
InternVL2-8B ↑1 0.17
- 8 MiniCPM-V-2_6 ↓4 0.17
- 9
LLaVA-v1.5-13B ↑3 0.17
- 10
LLaVA-NeXT-72B 0.16
HUST-OBS: Tier-3
15 models we track have a score here.
- 1 GPT-4o Aug 2024 0.37
- 2 GPT-4V ↑3 0.36
- 3
InternLM-XComposer2-VL-7B ↑6 0.27
- 4
InternVL2-Llama3-76B ↓2 0.22
- 5
mPLUG-Owl3-7B ↑9 0.19
- 6
InternVL2-40B ↓3 0.19
- 7
LLaVA-v1.5-13B ↑5 0.16
- 8 MiniCPM-V-2_6 ↓4 0.16
- 9
LLaVA-NeXT-72B ↑1 0.14
- 10
InternVL2-8B ↓2 0.13
EVOBC: pictograph
15 models we track have a score here.
- 1 GPT-4o Aug 2024 0.45
- 2 GPT-4V ↑3 0.44
- 3
InternLM-XComposer2-VL-7B ↑6 0.27
- 4
InternVL2-Llama3-76B ↓2 0.24
- 5
mPLUG-Owl3-7B ↑9 0.20
- 6
InternVL2-40B ↓3 0.20
- 7
InternVL2-8B ↑1 0.17
- 8 MiniCPM-V-2_6 ↓4 0.17
- 9
Idefics2-8B ↑6 0.16
- 10
LLaVA-NeXT-72B 0.15
EVOBC: ideogram
15 models we track have a score here.
- 1 GPT-4o Aug 2024 0.39
- 2 GPT-4V ↑3 0.36
- 3
Idefics2-8B ↑12 0.22
- 4
InternVL2-Llama3-76B ↓2 0.21
- 5
InternVL2-40B ↓2 0.19
- 6
InternLM-XComposer2-VL-7B ↑3 0.18
- 7
mPLUG-Owl3-7B ↑7 0.18
- 8 MiniCPM-V-2_6 ↓4 0.15
- 9
InternVL2-8B ↓1 0.15
- 10
LLaVA-v1.5-13B ↑2 0.12
OBI Component 20: Radical
15 models we track have a score here.
- 1 GPT-4V ↑4 0.41
- 2 GPT-4o Aug 2024 ↓1 0.38
- 3
InternVL2-Llama3-76B ↓1 0.19
- 4
InternVL2-8B ↑4 0.17
- 5
Idefics2-8B ↑10 0.15
- 6 llava-v1.5-7b ↑7 0.15
- 7
InternVL2-40B ↓4 0.14
- 8 CogVLM2-19B-Chat ↓1 0.12
- 9
LLaVA-v1.5-13B ↑3 0.12
- 10 MiniCPM-V-2_6 ↓6 0.09
Average
15 models we track have a score here.
- 1 GPT-4o Aug 2024 0.39
- 2 GPT-4V ↑3 0.39
- 3
InternLM-XComposer2-VL-7B ↑6 0.23
- 4
InternVL2-Llama3-76B ↓2 0.22
- 5
InternVL2-40B ↓2 0.18
- 6
mPLUG-Owl3-7B ↑8 0.16
- 7
InternVL2-8B ↑1 0.16
- 8
Idefics2-8B ↑7 0.15
- 9 MiniCPM-V-2_6 ↓5 0.15
- 10
LLaVA-v1.5-13B ↑2 0.14
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: The OBI-Bench paper
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Source: Official OBI-Bench leaderboard
No lab model card reports it yet
We read the headline results of 540 launch posts from 57 labs, and none of them list this benchmark.
More in Vision
- MMMU In use 11.5K college-level multimodal questions across 30 subjects. Multiple-choice and open accuracy.
Qwen 3.6 Plus 86%
- 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%
- MathVista In use Math reasoning in visual contexts (charts, geometry, puzzles) from 31 datasets. Answer accuracy.
Seed2.1 Pro 90.7%
- 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%
Data audit
8 of 9 checks pass, 1 warning. Look first at the Newest score check. The newest came 2 years ago, in August 2024, for mPLUG-Owl3-7B. Every benchmark
Catalog record 2 pass
- Pass Task access All tasks are public.
- Pass Confidence Checked against a primary source.
Scores 1 warning 2 pass 2 notes
- Pass Independent runs Scores from the benchmark’s paper (23 entries).
- Pass Models scored 15 models we track, plus 8 other entries on its leaderboard.
- Warning Newest score The newest came 2 years ago, in August 2024, for mPLUG-Owl3-7B.
- Note Frontier models None of the top 10 has a score. The strongest tested is GPT-4o (#167).
- Note Headroom Scored as a plain number, with no fixed ceiling.
Leaderboard import 3 pass
- Pass Importer We read the board on arxiv.org: 15 rows for models we track and 8 other entries at the last read.
- Pass Last read November 20th, 2026.
- Pass Read worked The last read worked.
Consistency 1 pass 1 note
- Note Lab claims No model has both a lab number and an independent run.
- Pass Known issues No published audit found problems.