LOKI
Models distinguish synthetic content from real content across images, videos, audio, text, and 3D data. The benchmark checks real-or-synthetic labels and annotated abnormal details, with separate tasks for explaining anomalies.
The suite spans 26 categories. Explanation tasks have different evaluation requirements from classification tasks.
- Made by
- OpenDataLab
- Released
- October 2024
- Kind
- General visual questions
- Level
- Everyday
- Scored
- Checked automatically
- Answers
- Multiple choice
- Tasks
- All public
- Measures
- Capability
- Language
- English
- Interaction
- Single turn
- Items from
- Generated, Real work and use
- Refresh
- One fixed set
- Run by
- Anyone
- In views
- Vision, Audio and speech
- Last model added
- August 2024
- Scores refreshed
- November 20th, 2026
In the authors’ words
With the rapid development of AI-generated content, the future internet may be inundated with synthetic data, making the discrimination of authentic and credible multimodal data increasingly challenging. Synthetic data detection has thus garnered widespread attention, and the performance of large multimodal models (LMMs) in this task has attracted significant interest. LMMs can provide natural language explanations for their authenticity judgments, enhancing the explainability of synthetic content detection. Simultaneously, the task of distinguishing between real and synthetic data effectively tests the perception, knowledge, and reasoning capabilities of LMMs. In response, we introduce LOKI, a novel benchmark designed to evaluate the ability of LMMs to detect synthetic data across multiple modalities. LOKI encompasses video, image, 3D, text, and audio modalities, comprising 18K carefully curated questions across 26 subcategories with clear difficulty levels. The benchmark includes coarse-grained judgment and multiple-choice questions, as well as fine-grained anomaly selection and explanation tasks, allowing for a comprehensive analysis of LMMs. We evaluated 22 open-source LMMs and 6 closed-source models on LOKI, highlighting their potential as synthetic data detectors and also revealing some limitations in the development of LMM capabilities. More information about LOKI can be found at https://opendatalab.github.io/LOKI/
GPT-4o (snapshot unspecified) tops the leaderboard with 68.8% LLaVA-OneVision 72B leads with 60.8% 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. This is the plain mean of the overall judgment and multiple choice accuracies in the multimodal evaluation. 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. The best model we track is LLaVA-OneVision 72B, at 60.8%. 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
- 68.8 GPT-4o (snapshot unspecified)
- 68.2 Claude-3.5-Sonnet (snapshot unspecified)
- 60.8 LLaVA-OneVision 72B
- 60.5 Qwen2-VL-72B-Instruct
- 58.0 Gemini-1.5-Pro (snapshot unspecified)
- 57.4 Qwen2-VL-7B-Instruct
- 55.2
LLaVA-OneVision 7B
- 54.0 Mantis-8B
- 53.2
VILA1.5-40B
- 52.5
InternVL2-40B
- 52.1 LongVA-7B
- 52.1
mPLUG-Owl3-7B
- 52.0
InternVL2-8B
- 51.5
InternVL2-26B
- 51.5 Phi-3.5-vision-instruct
- 51.3 MiniCPM-V-2_6
- 49.6
InternLM-XComposer2.5-7B
- 48.5
VILA1.5-13B
- 48.0 Llama-3-LongVILA-8B
- 47.9
Idefics2-8B
- 1 GPT-4o (snapshot unspecified) 68.8%
- 2 Claude-3.5-Sonnet (snapshot unspecified) 68.2%
- 3 LLaVA-OneVision 72B 60.8%
- 4 Qwen2-VL-72B-Instruct 60.5%
- 5 Gemini-1.5-Pro (snapshot unspecified) 58%
- 6 Qwen2-VL-7B-Instruct 57.4%
- 7
LLaVA-OneVision 7B 55.2%
- 8 Mantis-8B 54%
- 9
VILA1.5-40B 53.2%
- 10
InternVL2-40B 52.5%
- 11 LongVA-7B 52.1%
- 12
mPLUG-Owl3-7B 52.1%
- 13
InternVL2-8B 52%
- 14
InternVL2-26B 51.5%
- 15 Phi-3.5-vision-instruct 51.5%
- 16 MiniCPM-V-2_6 51.3%
- 17
InternLM-XComposer2.5-7B 49.6%
- 18
VILA1.5-13B 48.5%
- 19 Llama-3-LongVILA-8B 48%
- 20
Idefics2-8B 47.9%
Source: The LOKI paper
- 60.8 LLaVA-OneVision 72B
- 60.5 Qwen2-VL-72B-Instruct
- 57.4 Qwen2-VL-7B-Instruct
- 55.2
LLaVA-OneVision 7B
- 53.2
VILA1.5-40B
- 52.5
InternVL2-40B
- 52.1 LongVA-7B
- 52.1
mPLUG-Owl3-7B
- 52.0
InternVL2-8B
- 51.5
InternVL2-26B
- 51.5 Phi-3.5-vision-instruct
- 51.3 MiniCPM-V-2_6
- 49.6
InternLM-XComposer2.5-7B
- 48.5
VILA1.5-13B
- 47.9
Idefics2-8B
- 1 LLaVA-OneVision 72B 60.8%
- 2 Qwen2-VL-72B-Instruct 60.5%
- 3 Qwen2-VL-7B-Instruct 57.4%
- 4
LLaVA-OneVision 7B 55.2%
- 5
VILA1.5-40B 53.2%
- 6
InternVL2-40B 52.5%
- 7 LongVA-7B 52.1%
- 8
mPLUG-Owl3-7B 52.1%
- 9
InternVL2-8B 52%
- 10
InternVL2-26B 51.5%
- 11 Phi-3.5-vision-instruct 51.5%
- 12 MiniCPM-V-2_6 51.3%
- 13
InternLM-XComposer2.5-7B 49.6%
- 14
VILA1.5-13B 48.5%
- 15
Idefics2-8B 47.9%
Source: The LOKI paper
| # | Entry | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | GPT-4o (snapshot unspecified) | 68.8% | Official leaderboard | ||
| 2 | Claude-3.5-Sonnet (snapshot unspecified) | 68.2% | Official leaderboard | ||
| 3 | LLaVA-OneVision 72B | 60.8% | Aug 2024 | Official leaderboard | |
| 4 | Qwen2-VL-72B-Instruct | 60.5% | Aug 2024 | Official leaderboard | |
| 5 | Gemini-1.5-Pro (snapshot unspecified) | 58% | Official leaderboard | ||
| 6 | Qwen2-VL-7B-Instruct | 57.4% | Aug 2024 | Official leaderboard | |
| 7 | | 55.2% | Aug 2024 | Official leaderboard | |
| 8 | Mantis-8B | 54% | Official leaderboard | ||
| 9 | | 53.2% | Apr 2024 | Official leaderboard | |
| 10 | | 52.5% | Jul 2024 | Official leaderboard | |
| 11 | LongVA-7B | 52.1% | Jun 2024 | Official leaderboard | |
| 12 | | 52.1% | Aug 2024 | Official leaderboard | |
| 13 | | 52% | Jul 2024 | Official leaderboard | |
| 14 | | 51.5% | Jul 2024 | Official leaderboard | |
| 15 | Phi-3.5-vision-instruct | 51.5% | Aug 2024 | Official leaderboard | |
| 16 | MiniCPM-V-2_6 | 51.3% | Aug 2024 | Official leaderboard | |
| 17 | | 49.6% | Jul 2024 | Official leaderboard | |
| 18 | | 48.5% | Apr 2024 | Official leaderboard | |
| 19 | Llama-3-LongVILA-8B | 48% | Official leaderboard | ||
| 20 | | 47.9% | Apr 2024 | Official leaderboard |
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | LLaVA-OneVision 72B | 60.8% | Aug 2024 | Official leaderboard | |
| 2 | Qwen2-VL-72B-Instruct | 60.5% | Aug 2024 | Official leaderboard | |
| 3 | Qwen2-VL-7B-Instruct | 57.4% | Aug 2024 | Official leaderboard | |
| 4 | | 55.2% | Aug 2024 | Official leaderboard | |
| 5 | | 53.2% | Apr 2024 | Official leaderboard | |
| 6 | | 52.5% | Jul 2024 | Official leaderboard | |
| 7 | LongVA-7B | 52.1% | Jun 2024 | Official leaderboard | |
| 8 | | 52.1% | Aug 2024 | Official leaderboard | |
| 9 | | 52% | Jul 2024 | Official leaderboard | |
| 10 | | 51.5% | Jul 2024 | Official leaderboard | |
| 11 | Phi-3.5-vision-instruct | 51.5% | Aug 2024 | Official leaderboard | |
| 12 | MiniCPM-V-2_6 | 51.3% | Aug 2024 | Official leaderboard | |
| 13 | | 49.6% | Jul 2024 | Official leaderboard | |
| 14 | | 48.5% | Apr 2024 | Official leaderboard | |
| 15 | | 47.9% | Apr 2024 | Official leaderboard |
Rankings by part
We read 10 parts from the benchmark’s paper in 2 groups: judgment and multiple choice. Each ranks models on one part of the test. Qwen2-VL-72B-Instruct leads 3 of them, and 4 other models lead the rest. Arrows show how far a model moved from its overall rank among the models we track.
Video
15 models we track have a score here.
- 1
InternVL2-40B ↑5 62%
- 2
InternVL2-8B ↑7 60.8%
- 3 LongVA-7B ↑4 60.4%
- 4 Qwen2-VL-72B-Instruct ↓2 59.6%
- 5 Qwen2-VL-7B-Instruct ↓2 59.5%
- 6
VILA1.5-40B ↓1 59.2%
- 7
InternLM-XComposer2.5-7B ↑6 58.4%
- 8 MiniCPM-V-2_6 ↑4 57.2%
- 9
LLaVA-OneVision 7B ↓5 56.8%
- 10 Phi-3.5-vision-instruct ↑1 56.8%
Image
15 models we track have a score here.
- 1 Qwen2-VL-72B-Instruct ↑1 53.2%
- 2 Phi-3.5-vision-instruct ↑9 52.5%
- 3
LLaVA-OneVision 7B ↑1 49.8%
- 4
InternVL2-8B ↑5 49.7%
- 5
InternVL2-40B ↑1 49.6%
- 6
VILA1.5-13B ↑8 49.3%
- 7
VILA1.5-40B ↓2 48.8%
- 8 Qwen2-VL-7B-Instruct ↓5 47.8%
- 9
InternLM-XComposer2.5-7B ↑4 46.4%
- 10 LLaVA-OneVision 72B ↓9 46.3%
3D
15 models we track have a score here.
- 1 Qwen2-VL-7B-Instruct ↑2 72.3%
- 2
LLaVA-OneVision 7B ↑2 68.4%
- 3 Qwen2-VL-72B-Instruct ↓1 60.3%
- 4 MiniCPM-V-2_6 ↑8 56.4%
- 5 LLaVA-OneVision 72B ↓4 51.3%
- 6
InternVL2-26B ↑4 50.4%
- 7
VILA1.5-40B ↓2 50%
- 8 Phi-3.5-vision-instruct ↑3 50%
- 9 LongVA-7B ↓2 49.9%
- 10
InternVL2-40B ↓4 49.9%
Text
15 models we track have a score here.
- 1 LLaVA-OneVision 72B 61.2%
- 2
mPLUG-Owl3-7B ↑6 53.6%
- 3
InternVL2-40B ↑3 53.1%
- 4
LLaVA-OneVision 7B 53%
- 5 Qwen2-VL-72B-Instruct ↓3 52.8%
- 6
InternLM-XComposer2.5-7B ↑7 52.6%
- 7
InternVL2-26B ↑3 51.1%
- 8
InternVL2-8B ↑1 50.3%
- 9
VILA1.5-40B ↓4 50.1%
- 10 MiniCPM-V-2_6 ↑2 49.4%
Overall
15 models we track have a score here.
- 1 Qwen2-VL-7B-Instruct ↑2 57.1%
- 2
LLaVA-OneVision 7B ↑2 57%
- 3 LLaVA-OneVision 72B ↓2 56.3%
- 4 Qwen2-VL-72B-Instruct ↓2 55.4%
- 5
VILA1.5-40B 52.7%
- 6
InternVL2-8B ↑3 52.6%
- 7
InternVL2-40B ↓1 52.2%
- 8 Phi-3.5-vision-instruct ↑3 52.2%
- 9 MiniCPM-V-2_6 ↑3 52%
- 10 LongVA-7B ↓3 51.7%
Video
15 models we track have a score here.
- 1
InternVL2-40B ↑5 65.7%
- 2 Qwen2-VL-72B-Instruct 65.7%
- 3 Qwen2-VL-7B-Instruct 64%
- 4 LLaVA-OneVision 72B ↓3 62.9%
- 5
InternVL2-26B ↑5 62.4%
- 6
mPLUG-Owl3-7B ↑2 60.3%
- 7
LLaVA-OneVision 7B ↓3 59.8%
- 8 Phi-3.5-vision-instruct ↑3 58.2%
- 9 LongVA-7B ↓2 57.5%
- 10
InternLM-XComposer2.5-7B ↑3 56.3%
Image
15 models we track have a score here.
- 1 LLaVA-OneVision 72B 70.8%
- 2 Qwen2-VL-72B-Instruct 68.6%
- 3 Qwen2-VL-7B-Instruct 65.1%
- 4
VILA1.5-40B ↑1 64%
- 5
InternVL2-40B ↑1 63.1%
- 6
VILA1.5-13B ↑8 55.3%
- 7
mPLUG-Owl3-7B ↑1 52.5%
- 8
LLaVA-OneVision 7B ↓4 51.7%
- 9 LongVA-7B ↓2 51.6%
- 10
InternVL2-8B ↓1 51.4%
3D
15 models we track have a score here.
- 1 LongVA-7B ↑6 61.4%
- 2 LLaVA-OneVision 72B ↓1 61.3%
- 3
InternVL2-40B ↑3 59.9%
- 4 Phi-3.5-vision-instruct ↑7 59.6%
- 5 Qwen2-VL-72B-Instruct ↓3 58.7%
- 6 Qwen2-VL-7B-Instruct ↓3 55.5%
- 7
Idefics2-8B ↑8 54.2%
- 8
LLaVA-OneVision 7B ↓4 53.8%
- 9
VILA1.5-13B ↑5 53.5%
- 10
InternVL2-8B ↓1 53.1%
Text
15 models we track have a score here.
- 1 Qwen2-VL-72B-Instruct ↑1 69.7%
- 2 LLaVA-OneVision 72B ↓1 69.2%
- 3
VILA1.5-40B ↑2 50.4%
- 4
InternVL2-26B ↑6 50.3%
- 5
mPLUG-Owl3-7B ↑3 50%
- 6 LongVA-7B ↑1 48.9%
- 7 MiniCPM-V-2_6 ↑5 48.9%
- 8
LLaVA-OneVision 7B ↓4 48.4%
- 9
InternVL2-8B 46.6%
- 10 Qwen2-VL-7B-Instruct ↓7 46.4%
Overall
15 models we track have a score here.
- 1 Qwen2-VL-72B-Instruct ↑1 65.6%
- 2 LLaVA-OneVision 72B ↓1 65.2%
- 3 Qwen2-VL-7B-Instruct 57.7%
- 4
VILA1.5-40B ↑1 53.7%
- 5
LLaVA-OneVision 7B ↓1 53.4%
- 6
InternVL2-26B ↑4 53.2%
- 7
mPLUG-Owl3-7B ↑1 53.1%
- 8
InternVL2-40B ↓2 52.7%
- 9 LongVA-7B ↓2 52.6%
- 10
InternVL2-8B ↓1 51.3%
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: The LOKI 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 LOKI 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
9 of 10 checks pass, 1 warning. Look first at the Newest score check. The newest came 2 years ago, in August 2024, for Qwen2-VL-72B-Instruct. Every benchmark
Catalog record 2 pass
- Pass Task access All tasks are public.
- Pass Confidence Checked against a primary source.
Scores 1 warning 3 pass 1 note
- Pass Independent runs Scores from the benchmark’s paper (20 entries).
- Pass Models scored 15 models we track, plus 5 other entries on its leaderboard.
- Warning Newest score The newest came 2 years ago, in August 2024, for Qwen2-VL-72B-Instruct.
- Note Frontier models None of its models are on the capabilities index.
- Pass Headroom The best is 60.8%, under 85%, so it still separates models.
Leaderboard import 3 pass
- Pass Importer We read the board on arxiv.org: 15 rows for models we track and 5 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.