# Visual Hallucination Evaluation on HallusionBench

Source: [Visual Hallucination Evaluation on HallusionBench](<https://www.sota2.com/research/sota/visual-hallucination-evaluation-on-hallusionbench>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-07-07T07:41:18.664Z

Evaluation of model performance on visual hallucination detection using the HallusionBench dataset, typically measured by Accuracy.

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

| Scope | Recorded value |
| --- | --- |
| Task | Visual Hallucination Evaluation |
| Dataset (including any stated split) | HallusionBench |
| Metric | Accuracy |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 199 of 199 |
| Coverage | All results currently returned for this benchmark |

Evaluation settings may differ between results. The original papers and recorded configurations are needed to establish comparability. A shared metric name alone does not establish an identical protocol.

## Metrics

- [Accuracy](<https://www.sota2.com/research/sota/visual-hallucination-evaluation-on-hallusionbench.md?metric=Accuracy>): Higher is better; unit: PERCENT.

- [Accuracy (Q)](<https://www.sota2.com/research/sota/visual-hallucination-evaluation-on-hallusionbench.md?metric=Accuracy%20(Q)>): Higher is better; unit: PERCENT.

- [Accuracy (F)](<https://www.sota2.com/research/sota/visual-hallucination-evaluation-on-hallusionbench.md?metric=Accuracy%20(F)>): Higher is better; unit: PERCENT.

- [Accuracy (Avg)](<https://www.sota2.com/research/sota/visual-hallucination-evaluation-on-hallusionbench.md?metric=Accuracy%20(Avg)>): Higher is better; unit: PERCENT.

- [Average Tokens](<https://www.sota2.com/research/sota/visual-hallucination-evaluation-on-hallusionbench.md?metric=Average%20Tokens>): Higher is better; unit: SCALAR.

## Recorded results

| Method | Accuracy | Result date | Report paper |
| --- | --- | --- | --- |
| Qwen2-VL-72B-Thinking | 76.6 | Not recorded | 69a90e162d603e6de3e408e6 |
| Qwen2-VL-72B-Thinking | 76.4 | Not recorded | 69a90e162d603e6de3e408e6 |
| Vision-SR1 | 75.6 | Not recorded | 68fa7ae11e209131cea85a1c |
| InternVL3.5-38B | 75.07 | Not recorded | 6a4c87fa85dba69fef8b2578 |
| Qwen3-VL + DISCO (10K) | 74.97 | Not recorded | 6982d2b1a8b01e76cc8e95fb |
| Vision-R1 | 74.3 | Not recorded | 68fa7ae11e209131cea85a1c |
| Visual Para-Thinker++ | 73.8 | Not recorded | 6a279e5b90966ade919bfa4c |
| Qwen3-VL | 73.5 | Not recorded | 6982d2b1a8b01e76cc8e95fb |
| Qwen2-VL-7B-Thinking | 73 | Not recorded | 69a90e162d603e6de3e408e6 |
| Qwen2.5-VL-7B + ECRD | 72.5 | Not recorded | 699fd36a2d603e6de3e3bda3 |
| Qwen3-VL + Textual (10K) | 72.24 | Not recorded | 6982d2b1a8b01e76cc8e95fb |
| Zero-shot Inference | 71.9 | Not recorded | 68fa7ae11e209131cea85a1c |
| Qwen2-VL-7B-Thinking | 71.6 | Not recorded | 69a90e162d603e6de3e408e6 |
| InternVL3.5-8B | 71.39 | Not recorded | 6a4c87fa85dba69fef8b2578 |
| Visual Para-Thinker | 71 | Not recorded | 6a279e5b90966ade919bfa4c |
| Kimi-VL-A3B-Thinking | 70.6 | Not recorded | 69a90e162d603e6de3e408e6 |
| Qwen2.5-VL | 70 | Not recorded | 6a279e5b90966ade919bfa4c |
| Qwen2.5-VL-72B-Instruct | 69.61 | Not recorded | 6a4c87fa85dba69fef8b2578 |
| Optimization-based Token Selection | 69.6 | Not recorded | 6a4c87fa85dba69fef8b2578 |
| InternVL3.5-4B | 69.08 | Not recorded | 6a4c87fa85dba69fef8b2578 |
| Vision-SR1 | 68.9 | Not recorded | 68fa7ae11e209131cea85a1c |
| GRPO | 68.8 | Not recorded | 6a279e5b90966ade919bfa4c |
| Vision-SR1 | 68.3 | Not recorded | 68fa7ae11e209131cea85a1c |
| Qwen2.5-VL-7B + supervisor | 67.5 | Not recorded | 699fd36a2d603e6de3e3bda3 |
| Vision-R1 | 67.4 | Not recorded | 68fa7ae11e209131cea85a1c |
| Majority voting@4 | 66.8 | Not recorded | 6a279e5b90966ade919bfa4c |
| Sequential (SFT) | 66.7 | Not recorded | 6a279e5b90966ade919bfa4c |
| Vision-R1 | 66.6 | Not recorded | 68fa7ae11e209131cea85a1c |
| Qwen2.5-VL | 66 | Not recorded | 6a279e5b90966ade919bfa4c |
| Percention-R1 (7B) | 65.4 | Not recorded | 68fa7ae11e209131cea85a1c |
| gemma-2-9b-it | 65.3 | Not recorded | 69a90e162d603e6de3e408e6 |
| Qwen2.5-VL-7B-Instruct | 65.19 | Not recorded | 6a4c87fa85dba69fef8b2578 |
| Visual Para-Thinker++ | 64.7 | Not recorded | 6a279e5b90966ade919bfa4c |
| Phi-4-reasoning-vision-15B | 64.4 | Not recorded | 69a90e162d603e6de3e408e6 |
| Visual Para-Thinker++ | 64 | Not recorded | 6a279e5b90966ade919bfa4c |
| Phi-4-reasoning-vision-15B | 63.9 | Not recorded | 69a90e162d603e6de3e408e6 |
| MiniCPM-v2.6 | 63.83 | Not recorded | 6982d2b1a8b01e76cc8e95fb |
| LLaVA-OneVision-7B + ECRD | 63.7 | Not recorded | 699fd36a2d603e6de3e3bda3 |
| Gemini2.5-Pro | 63.7 | Not recorded | 6a279e5b90966ade919bfa4c |
| Qwen2.5-VL-3B-Instruct | 63.62 | Not recorded | 692fc434d6b0dd3342933465 |
| MiniCPM-v2.6 + HIPPO | 63.41 | Not recorded | 6982d2b1a8b01e76cc8e95fb |
| GPT5-mini | 63.2 | Not recorded | 6a279e5b90966ade919bfa4c |
| POW3R dynamic | 62.9 | Not recorded | 6a0d405ef984739ed6737bcf |
| Category-balanced | 62.6 | Not recorded | 6a0d405ef984739ed6737bcf |
| Category-balanced | 62.5 | Not recorded | 6a0d405ef984739ed6737bcf |
| Binary | 62.5 | Not recorded | 6a0d405ef984739ed6737bcf |
| POW3R dynamic | 62.4 | Not recorded | 6a0d405ef984739ed6737bcf |
| Base | 62.4 | Not recorded | 6a0d405ef984739ed6737bcf |
| Static scalar | 62.2 | Not recorded | 6a0d405ef984739ed6737bcf |
| Static scalar | 62.2 | Not recorded | 6a0d405ef984739ed6737bcf |
| Visual Para-Thinker | 62.2 | Not recorded | 6a279e5b90966ade919bfa4c |
| Base | 62.1 | Not recorded | 6a0d405ef984739ed6737bcf |
| Qwen2-VL-2B | 62.04 | Not recorded | 692fc434d6b0dd3342933465 |
| Binary | 62 | Not recorded | 6a0d405ef984739ed6737bcf |
| EVE-iter3 | 61.89 | Not recorded | 69e705217d0172140ec7f2bd |
| GPT-4o | 61.4 | Not recorded | 6a279e5b90966ade919bfa4c |
| Qwen2.5-VL-7B | 61.3 | Not recorded | 699fd36a2d603e6de3e3bda3 |
| EVE-iter1 | 61.28 | Not recorded | 69e705217d0172140ec7f2bd |
| EVE-iter2 | 61.09 | Not recorded | 69e705217d0172140ec7f2bd |
| MiMo-VL-7B-SFT-2508 | 60.93 | Not recorded | 69e705217d0172140ec7f2bd |
| GRPO | 60.4 | Not recorded | 6a279e5b90966ade919bfa4c |
| Qwen2.5-VL-3B-Instruct | 60.04 | Not recorded | 6a4c87fa85dba69fef8b2578 |
| Qwen3-VL-8B | 59.54 | Not recorded | 6a0d405ff984739ed6737bdc |
| Claude-4-Sonnet | 59.2 | Not recorded | 6a279e5b90966ade919bfa4c |
| InternVL3.5-38B | 59.08 | Not recorded | 6a0bee84dbc836b87d82c339 |
| F3A | 58.9 | Not recorded | 6a0bee84dbc836b87d82c339 |
| VisionZip | 58.77 | Not recorded | 6a0bee84dbc836b87d82c339 |
| DivPrune | 58.71 | Not recorded | 6a0bee84dbc836b87d82c339 |
| FastV | 58.38 | Not recorded | 6a0bee84dbc836b87d82c339 |
| InternVL2-2B | 58.36 | Not recorded | 692fc434d6b0dd3342933465 |
| VisionZip | 58.35 | Not recorded | 6a0bee84dbc836b87d82c339 |
| Sequential (SFT) | 58.3 | Not recorded | 6a279e5b90966ade919bfa4c |
| Majority voting@4 | 58.1 | Not recorded | 6a279e5b90966ade919bfa4c |
| Visual Para-Thinker++ | 58 | Not recorded | 6a279e5b90966ade919bfa4c |
| F3A | 57.96 | Not recorded | 6a0bee84dbc836b87d82c339 |
| MobileNet-Qwen | 57.94 | Not recorded | 692fc434d6b0dd3342933465 |
| DivPrune | 57.78 | Not recorded | 6a0bee84dbc836b87d82c339 |
| FastV | 57.31 | Not recorded | 6a0bee84dbc836b87d82c339 |
| Honeybee-Remake-SEED-200K | 56.8 | Not recorded | 6a0a9cdedbc836b87d82b23f |
| BASE | 56.7 | Not recorded | 6a0a9cdedbc836b87d82b23f |

### Qwen2-VL-72B-Thinking

Accuracy: 76.6. Result ID: 69a948786dbb7075b952f6ea.

[Reporting paper](<https://www.sota2.com/research/paper/69a90e162d603e6de3e408e6.md>)

Recorded evaluation configuration: {"Max Output Tokens":"40K"}. This record does not assert independent reproduction.

### Qwen2-VL-72B-Thinking

Accuracy: 76.4. Result ID: 69a948786dbb7075b952f6e9.

[Reporting paper](<https://www.sota2.com/research/paper/69a90e162d603e6de3e408e6.md>)

Recorded evaluation configuration: {"Max Output Tokens":"4K"}. This record does not assert independent reproduction.

### Vision-SR1

Accuracy: 75.6. Result ID: 69f0836a47f34ce924d6eab0.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7ae11e209131cea85a1c.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7ae11e209131cea85a1c.md>)

Recorded evaluation configuration: {"Backbone":"Mimo-VL-7B"}. This record does not assert independent reproduction.

### InternVL3.5-38B

Accuracy: 75.07. Result ID: 6a4cad8260b88bb339c7bfc5.

[Reporting paper](<https://www.sota2.com/research/paper/6a4c87fa85dba69fef8b2578.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7af01e209131cea85dc9.md>)

Recorded evaluation configuration: {"Attack":"Benign"}. This record does not assert independent reproduction.

### Qwen3-VL + DISCO (10K)

Accuracy: 74.97. Result ID: 69974dcd5698665ec49f5981.

[Reporting paper](<https://www.sota2.com/research/paper/6982d2b1a8b01e76cc8e95fb.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6982d2b1a8b01e76cc8e95fb.md>)

Recorded evaluation configuration: {"Backbone":"Qwen3-VL","Alignment Strategy":"DISCO (10K)"}. This record does not assert independent reproduction.

### Vision-R1

Accuracy: 74.3. Result ID: 69f0836a47f34ce924d6eaaf.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7ae11e209131cea85a1c.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa77591e209131cea72979.md>)

Recorded evaluation configuration: {"Backbone":"Mimo-VL-7B","Training Data":"47K data"}. This record does not assert independent reproduction.

### Visual Para-Thinker++

Accuracy: 73.8. Result ID: 6a27ed5cfbc373592c257221.

[Reporting paper](<https://www.sota2.com/research/paper/6a279e5b90966ade919bfa4c.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a279e5b90966ade919bfa4c.md>)

Recorded evaluation configuration: {"Size":"7B"}. This record does not assert independent reproduction.

### Qwen3-VL

Accuracy: 73.5. Result ID: 69974dcd5698665ec49f597f.

[Reporting paper](<https://www.sota2.com/research/paper/6982d2b1a8b01e76cc8e95fb.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6927db54d6b0dd3342930f66.md>)

Recorded evaluation configuration: {"Backbone":"Qwen3-VL"}. This record does not assert independent reproduction.

### Qwen2-VL-7B-Thinking

Accuracy: 73. Result ID: 69a948786dbb7075b952f6e8.

[Reporting paper](<https://www.sota2.com/research/paper/69a90e162d603e6de3e408e6.md>)

Recorded evaluation configuration: {"Max Output Tokens":"40K"}. This record does not assert independent reproduction.

### Qwen2.5-VL-7B + ECRD

Accuracy: 72.5. Result ID: 69afc9bf6dbb7075b957cb51.

[Reporting paper](<https://www.sota2.com/research/paper/699fd36a2d603e6de3e3bda3.md>)

[Method origin paper](<https://www.sota2.com/research/paper/699fd36a2d603e6de3e3bda3.md>)

Recorded evaluation configuration: {"Parameters":"7B","Enhancement":"ECRD"}. This record does not assert independent reproduction.

### Qwen3-VL + Textual (10K)

Accuracy: 72.24. Result ID: 69974dcd5698665ec49f5980.

[Reporting paper](<https://www.sota2.com/research/paper/6982d2b1a8b01e76cc8e95fb.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6927db54d6b0dd3342930f66.md>)

Recorded evaluation configuration: {"Backbone":"Qwen3-VL","Alignment Strategy":"Textual (10K)"}. This record does not assert independent reproduction.

### Zero-shot Inference

Accuracy: 71.9. Result ID: 69f0836a47f34ce924d6eaae.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7ae11e209131cea85a1c.md>)

Recorded evaluation configuration: {"Backbone":"Mimo-VL-7B","Inference Protocol":"Zero-shot (before RL)"}. This record does not assert independent reproduction.

### Qwen2-VL-7B-Thinking

Accuracy: 71.6. Result ID: 69a948786dbb7075b952f6e7.

[Reporting paper](<https://www.sota2.com/research/paper/69a90e162d603e6de3e408e6.md>)

Recorded evaluation configuration: {"Max Output Tokens":"4K"}. This record does not assert independent reproduction.

### InternVL3.5-8B

Accuracy: 71.39. Result ID: 6a4cad8260b88bb339c7bfc3.

[Reporting paper](<https://www.sota2.com/research/paper/6a4c87fa85dba69fef8b2578.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7af01e209131cea85dc9.md>)

Recorded evaluation configuration: {"Attack":"Benign"}. This record does not assert independent reproduction.

### Visual Para-Thinker

Accuracy: 71. Result ID: 6a27ed5cfbc373592c25721f.

[Reporting paper](<https://www.sota2.com/research/paper/6a279e5b90966ade919bfa4c.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7b1b1e209131cea86980.md>)

Recorded evaluation configuration: {"Size":"7B"}. This record does not assert independent reproduction.

### Kimi-VL-A3B-Thinking

Accuracy: 70.6. Result ID: 69a948786dbb7075b952f6e5.

[Reporting paper](<https://www.sota2.com/research/paper/69a90e162d603e6de3e408e6.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa79a81e209131cea7eb7f.md>)

Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction.

### Qwen2.5-VL

Accuracy: 70. Result ID: 6a27ed5cfbc373592c257215.

[Reporting paper](<https://www.sota2.com/research/paper/6a279e5b90966ade919bfa4c.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa773a1e209131cea71d3c.md>)

Recorded evaluation configuration: {"Size":"72B"}. This record does not assert independent reproduction.

### Qwen2.5-VL-72B-Instruct

Accuracy: 69.61. Result ID: 6a4cad8260b88bb339c7bfbf.

[Reporting paper](<https://www.sota2.com/research/paper/6a4c87fa85dba69fef8b2578.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa773a1e209131cea71d3c.md>)

Recorded evaluation configuration: {"Attack":"Benign"}. This record does not assert independent reproduction.

### Optimization-based Token Selection

Accuracy: 69.6. Result ID: 6a4cad8260b88bb339c7bfc4.

[Reporting paper](<https://www.sota2.com/research/paper/6a4c87fa85dba69fef8b2578.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a4c87fa85dba69fef8b2578.md>)

Recorded evaluation configuration: {"Target Model":"InternVL3.5-8B"}. This record does not assert independent reproduction.

### InternVL3.5-4B

Accuracy: 69.08. Result ID: 6a4cad8260b88bb339c7bfc1.

[Reporting paper](<https://www.sota2.com/research/paper/6a4c87fa85dba69fef8b2578.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7af01e209131cea85dc9.md>)

Recorded evaluation configuration: {"Attack":"Benign"}. This record does not assert independent reproduction.

### Vision-SR1

Accuracy: 68.9. Result ID: 69f0836a47f34ce924d6eaad.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7ae11e209131cea85a1c.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7ae11e209131cea85a1c.md>)

Recorded evaluation configuration: {"Backbone":"Qwen2.5-VL-7B"}. This record does not assert independent reproduction.

### GRPO

Accuracy: 68.8. Result ID: 6a27ed5cfbc373592c25721b.

[Reporting paper](<https://www.sota2.com/research/paper/6a279e5b90966ade919bfa4c.md>)

Recorded evaluation configuration: {"Size":"7B"}. This record does not assert independent reproduction.

### Vision-SR1

Accuracy: 68.3. Result ID: 69f0836a47f34ce924d6eaaa.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7ae11e209131cea85a1c.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7ae11e209131cea85a1c.md>)

Recorded evaluation configuration: {"Backbone":"Qwen2.5-VL-3B"}. This record does not assert independent reproduction.

### Qwen2.5-VL-7B + supervisor

Accuracy: 67.5. Result ID: 69afc9bf6dbb7075b957cb50.

[Reporting paper](<https://www.sota2.com/research/paper/699fd36a2d603e6de3e3bda3.md>)

[Method origin paper](<https://www.sota2.com/research/paper/699fd36a2d603e6de3e3bda3.md>)

Recorded evaluation configuration: {"Parameters":"7B","Enhancement":"supervisor"}. This record does not assert independent reproduction.

### Vision-R1

Accuracy: 67.4. Result ID: 69f0836a47f34ce924d6eaa9.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7ae11e209131cea85a1c.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa77591e209131cea72979.md>)

Recorded evaluation configuration: {"Backbone":"Qwen2.5-VL-3B","Training Data":"47K data"}. This record does not assert independent reproduction.

### Majority voting@4

Accuracy: 66.8. Result ID: 6a27ed5cfbc373592c25721d.

[Reporting paper](<https://www.sota2.com/research/paper/6a279e5b90966ade919bfa4c.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6eb61e209131cea41b33.md>)

Recorded evaluation configuration: {"Size":"7B"}. This record does not assert independent reproduction.

### Sequential (SFT)

Accuracy: 66.7. Result ID: 6a27ed5cfbc373592c257219.

[Reporting paper](<https://www.sota2.com/research/paper/6a279e5b90966ade919bfa4c.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7b1b1e209131cea86980.md>)

Recorded evaluation configuration: {"Size":"7B"}. This record does not assert independent reproduction.

### Vision-R1

Accuracy: 66.6. Result ID: 69f0836a47f34ce924d6eaac.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7ae11e209131cea85a1c.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa77591e209131cea72979.md>)

Recorded evaluation configuration: {"Backbone":"Qwen2.5-VL-7B","Training Data":"47K data"}. This record does not assert independent reproduction.

### Qwen2.5-VL

Accuracy: 66. Result ID: 6a27ed5cfbc373592c257217.

[Reporting paper](<https://www.sota2.com/research/paper/6a279e5b90966ade919bfa4c.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa773a1e209131cea71d3c.md>)

Recorded evaluation configuration: {"Size":"7B"}. This record does not assert independent reproduction.

### Percention-R1 (7B)

Accuracy: 65.4. Result ID: 69f0836a47f34ce924d6eaa6.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7ae11e209131cea85a1c.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7b651e209131cea882b4.md>)

Recorded evaluation configuration: {"Backbone":"7B"}. This record does not assert independent reproduction.

### gemma-2-9b-it

Accuracy: 65.3. Result ID: 69a948786dbb7075b952f6e6.

[Reporting paper](<https://www.sota2.com/research/paper/69a90e162d603e6de3e408e6.md>)

Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction.

### Qwen2.5-VL-7B-Instruct

Accuracy: 65.19. Result ID: 6a4cad8260b88bb339c7bfbd.

[Reporting paper](<https://www.sota2.com/research/paper/6a4c87fa85dba69fef8b2578.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa773a1e209131cea71d3c.md>)

Recorded evaluation configuration: {"Attack":"Benign"}. This record does not assert independent reproduction.

### Visual Para-Thinker++

Accuracy: 64.7. Result ID: 6a27ed5cfbc373592c257270.

[Reporting paper](<https://www.sota2.com/research/paper/6a279e5b90966ade919bfa4c.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a279e5b90966ade919bfa4c.md>)

Recorded evaluation configuration: {"Backbone":"Qwen2.5-VL-7B","Training Stage (MAC/RDMAO)":"MAC+RDMAO"}. This record does not assert independent reproduction.

### Phi-4-reasoning-vision-15B

Accuracy: 64.4. Result ID: 69a948786dbb7075b952f6e3.

[Reporting paper](<https://www.sota2.com/research/paper/69a90e162d603e6de3e408e6.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69a90e162d603e6de3e408e6.md>)

[Implementation](<https://github.com/microsoft/Phi-4-reasoning-vision-15B>)

Recorded evaluation configuration: {"Thinking Mode":"default"}. This record does not assert independent reproduction.

### Visual Para-Thinker++

Accuracy: 64. Result ID: 6a27ed5cfbc373592c257220.

[Reporting paper](<https://www.sota2.com/research/paper/6a279e5b90966ade919bfa4c.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a279e5b90966ade919bfa4c.md>)

Recorded evaluation configuration: {"Size":"3B"}. This record does not assert independent reproduction.

### Phi-4-reasoning-vision-15B

Accuracy: 63.9. Result ID: 69a948786dbb7075b952f6e4.

[Reporting paper](<https://www.sota2.com/research/paper/69a90e162d603e6de3e408e6.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69a90e162d603e6de3e408e6.md>)

[Implementation](<https://github.com/microsoft/Phi-4-reasoning-vision-15B>)

Recorded evaluation configuration: {"Thinking Mode":"force thinking"}. This record does not assert independent reproduction.

### MiniCPM-v2.6

Accuracy: 63.83. Result ID: 69974dcd5698665ec49f597d.

[Reporting paper](<https://www.sota2.com/research/paper/6982d2b1a8b01e76cc8e95fb.md>)

Recorded evaluation configuration: {"Backbone":"MiniCPM-v2.6"}. This record does not assert independent reproduction.

### LLaVA-OneVision-7B + ECRD

Accuracy: 63.7. Result ID: 69afc9bf6dbb7075b957cb4e.

[Reporting paper](<https://www.sota2.com/research/paper/699fd36a2d603e6de3e3bda3.md>)

[Method origin paper](<https://www.sota2.com/research/paper/699fd36a2d603e6de3e3bda3.md>)

Recorded evaluation configuration: {"Parameters":"7B","Enhancement":"ECRD"}. This record does not assert independent reproduction.

### Gemini2.5-Pro

Accuracy: 63.7. Result ID: 6a27ed5cfbc373592c257213.

[Reporting paper](<https://www.sota2.com/research/paper/6a279e5b90966ade919bfa4c.md>)

Recorded evaluation configuration: {"Size":"-"}. This record does not assert independent reproduction.

### Qwen2.5-VL-3B-Instruct

Accuracy: 63.62. Result ID: 6996d0735698665ec41c8cf5.

[Reporting paper](<https://www.sota2.com/research/paper/692fc434d6b0dd3342933465.md>)

Recorded evaluation configuration: {"Params":"3.75B"}. This record does not assert independent reproduction.

### MiniCPM-v2.6 + HIPPO

Accuracy: 63.41. Result ID: 69974dcd5698665ec49f597e.

[Reporting paper](<https://www.sota2.com/research/paper/6982d2b1a8b01e76cc8e95fb.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa771a1e209131cea71018.md>)

Recorded evaluation configuration: {"Backbone":"MiniCPM-v2.6","Alignment Strategy":"HIPPO"}. This record does not assert independent reproduction.

### GPT5-mini

Accuracy: 63.2. Result ID: 6a27ed5cfbc373592c257211.

[Reporting paper](<https://www.sota2.com/research/paper/6a279e5b90966ade919bfa4c.md>)

Recorded evaluation configuration: {"Size":"-"}. This record does not assert independent reproduction.

### POW3R dynamic

Accuracy: 62.9. Result ID: 6a0d83d8fbc373592c06049d.

[Reporting paper](<https://www.sota2.com/research/paper/6a0d405ef984739ed6737bcf.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a0d405ef984739ed6737bcf.md>)

Recorded evaluation configuration: {"Base policy":"Qwen3-VL-8B"}. This record does not assert independent reproduction.

### Category-balanced

Accuracy: 62.6. Result ID: 6a0d83d8fbc373592c06049c.

[Reporting paper](<https://www.sota2.com/research/paper/6a0d405ef984739ed6737bcf.md>)

Recorded evaluation configuration: {"Base policy":"Qwen3-VL-8B"}. This record does not assert independent reproduction.

### Category-balanced

Accuracy: 62.5. Result ID: 6a0d83d8fbc373592c060497.

[Reporting paper](<https://www.sota2.com/research/paper/6a0d405ef984739ed6737bcf.md>)

Recorded evaluation configuration: {"Base policy":"Qwen3-VL-4B"}. This record does not assert independent reproduction.

### Binary

Accuracy: 62.5. Result ID: 6a0d83d8fbc373592c06049a.

[Reporting paper](<https://www.sota2.com/research/paper/6a0d405ef984739ed6737bcf.md>)

Recorded evaluation configuration: {"Base policy":"Qwen3-VL-8B"}. This record does not assert independent reproduction.

### POW3R dynamic

Accuracy: 62.4. Result ID: 6a0d83d8fbc373592c060498.

[Reporting paper](<https://www.sota2.com/research/paper/6a0d405ef984739ed6737bcf.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a0d405ef984739ed6737bcf.md>)

Recorded evaluation configuration: {"Base policy":"Qwen3-VL-4B"}. This record does not assert independent reproduction.

### Base

Accuracy: 62.4. Result ID: 6a0d83d8fbc373592c060499.

[Reporting paper](<https://www.sota2.com/research/paper/6a0d405ef984739ed6737bcf.md>)

Recorded evaluation configuration: {"Base policy":"Qwen3-VL-8B"}. This record does not assert independent reproduction.

### Static scalar

Accuracy: 62.2. Result ID: 6a0d83d8fbc373592c060496.

[Reporting paper](<https://www.sota2.com/research/paper/6a0d405ef984739ed6737bcf.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7bce1e209131cea8a8ec.md>)

Recorded evaluation configuration: {"Base policy":"Qwen3-VL-4B"}. This record does not assert independent reproduction.

### Static scalar

Accuracy: 62.2. Result ID: 6a0d83d8fbc373592c06049b.

[Reporting paper](<https://www.sota2.com/research/paper/6a0d405ef984739ed6737bcf.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7bce1e209131cea8a8ec.md>)

Recorded evaluation configuration: {"Base policy":"Qwen3-VL-8B"}. This record does not assert independent reproduction.

### Visual Para-Thinker

Accuracy: 62.2. Result ID: 6a27ed5cfbc373592c25721e.

[Reporting paper](<https://www.sota2.com/research/paper/6a279e5b90966ade919bfa4c.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7b1b1e209131cea86980.md>)

Recorded evaluation configuration: {"Size":"3B"}. This record does not assert independent reproduction.

### Base

Accuracy: 62.1. Result ID: 6a0d83d8fbc373592c060494.

[Reporting paper](<https://www.sota2.com/research/paper/6a0d405ef984739ed6737bcf.md>)

Recorded evaluation configuration: {"Base policy":"Qwen3-VL-4B"}. This record does not assert independent reproduction.

### Qwen2-VL-2B

Accuracy: 62.04. Result ID: 6996d0735698665ec41c8cf4.

[Reporting paper](<https://www.sota2.com/research/paper/692fc434d6b0dd3342933465.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa75151e209131cea65976.md>)

[Implementation](<https://github.com/QwenLM/Qwen2-VL>)

Recorded evaluation configuration: {"Params":"2.2B"}. This record does not assert independent reproduction.

### Binary

Accuracy: 62. Result ID: 6a0d83d8fbc373592c060495.

[Reporting paper](<https://www.sota2.com/research/paper/6a0d405ef984739ed6737bcf.md>)

Recorded evaluation configuration: {"Base policy":"Qwen3-VL-4B"}. This record does not assert independent reproduction.

### EVE-iter3

Accuracy: 61.89. Result ID: 69e778e547f34ce924cf555b.

[Reporting paper](<https://www.sota2.com/research/paper/69e705217d0172140ec7f2bd.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69e705217d0172140ec7f2bd.md>)

Recorded evaluation configuration: {"Backbone":"MiMo-VL-7B-SFT-2508","Training Iteration":"3"}. This record does not assert independent reproduction.

### GPT-4o

Accuracy: 61.4. Result ID: 6a27ed5cfbc373592c257212.

[Reporting paper](<https://www.sota2.com/research/paper/6a279e5b90966ade919bfa4c.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa75681e209131cea67a98.md>)

Recorded evaluation configuration: {"Size":"-"}. This record does not assert independent reproduction.

### Qwen2.5-VL-7B

Accuracy: 61.3. Result ID: 69afc9bf6dbb7075b957cb4f.

[Reporting paper](<https://www.sota2.com/research/paper/699fd36a2d603e6de3e3bda3.md>)

[Method origin paper](<https://www.sota2.com/research/paper/699fd36a2d603e6de3e3bda3.md>)

Recorded evaluation configuration: {"Parameters":"7B"}. This record does not assert independent reproduction.

### EVE-iter1

Accuracy: 61.28. Result ID: 69e778e547f34ce924cf5559.

[Reporting paper](<https://www.sota2.com/research/paper/69e705217d0172140ec7f2bd.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69e705217d0172140ec7f2bd.md>)

Recorded evaluation configuration: {"Backbone":"MiMo-VL-7B-SFT-2508","Training Iteration":"1"}. This record does not assert independent reproduction.

### EVE-iter2

Accuracy: 61.09. Result ID: 69e778e547f34ce924cf555a.

[Reporting paper](<https://www.sota2.com/research/paper/69e705217d0172140ec7f2bd.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69e705217d0172140ec7f2bd.md>)

Recorded evaluation configuration: {"Backbone":"MiMo-VL-7B-SFT-2508","Training Iteration":"2"}. This record does not assert independent reproduction.

### MiMo-VL-7B-SFT-2508

Accuracy: 60.93. Result ID: 69e778e547f34ce924cf5558.

[Reporting paper](<https://www.sota2.com/research/paper/69e705217d0172140ec7f2bd.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa79121e209131cea7bfbd.md>)

Recorded evaluation configuration: {"Backbone":"MiMo-VL-7B-SFT-2508","Training Iteration":"0"}. This record does not assert independent reproduction.

### GRPO

Accuracy: 60.4. Result ID: 6a27ed5cfbc373592c25721a.

[Reporting paper](<https://www.sota2.com/research/paper/6a279e5b90966ade919bfa4c.md>)

Recorded evaluation configuration: {"Size":"3B"}. This record does not assert independent reproduction.

### Qwen2.5-VL-3B-Instruct

Accuracy: 60.04. Result ID: 6a4cad8260b88bb339c7bfbb.

[Reporting paper](<https://www.sota2.com/research/paper/6a4c87fa85dba69fef8b2578.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa773a1e209131cea71d3c.md>)

Recorded evaluation configuration: {"Attack":"Benign"}. This record does not assert independent reproduction.

### Qwen3-VL-8B

Accuracy: 59.54. Result ID: 6a0d846ffbc373592c0606d5.

[Reporting paper](<https://www.sota2.com/research/paper/6a0d405ff984739ed6737bdc.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a0d405ff984739ed6737bdc.md>)

[Implementation](<https://github.com/UCSC-VLAA/VLM-CapCurriculum>)

Recorded evaluation configuration: {"Training Strategy":"Staged"}. This record does not assert independent reproduction.

### Claude-4-Sonnet

Accuracy: 59.2. Result ID: 6a27ed5cfbc373592c257214.

[Reporting paper](<https://www.sota2.com/research/paper/6a279e5b90966ade919bfa4c.md>)

Recorded evaluation configuration: {"Size":"-"}. This record does not assert independent reproduction.

### InternVL3.5-38B

Accuracy: 59.08. Result ID: 6a0c13f2fbc373592c01d115.

[Reporting paper](<https://www.sota2.com/research/paper/6a0bee84dbc836b87d82c339.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa71de1e209131cea5370d.md>)

[Implementation](<https://github.com/OpenGVLab/InternVL>)

Recorded evaluation configuration: {"Ratio":"100%","Backbone":"InternVL3.5-38B"}. This record does not assert independent reproduction.

### F3A

Accuracy: 58.9. Result ID: 6a0c13f2fbc373592c01d11a.

[Reporting paper](<https://www.sota2.com/research/paper/6a0bee84dbc836b87d82c339.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a0bee84dbc836b87d82c339.md>)

Recorded evaluation configuration: {"Ratio":"60%","Backbone":"InternVL3.5-38B"}. This record does not assert independent reproduction.

### VisionZip

Accuracy: 58.77. Result ID: 6a0c13f2fbc373592c01d11e.

[Reporting paper](<https://www.sota2.com/research/paper/6a0bee84dbc836b87d82c339.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa760f1e209131cea6b33b.md>)

[Implementation](<https://github.com/dvlab-research/VisionZip>)

Recorded evaluation configuration: {"Ratio":"40%","Backbone":"InternVL3.5-38B"}. This record does not assert independent reproduction.

### DivPrune

Accuracy: 58.71. Result ID: 6a0c13f2fbc373592c01d118.

[Reporting paper](<https://www.sota2.com/research/paper/6a0bee84dbc836b87d82c339.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa77c41e209131cea7532a.md>)

Recorded evaluation configuration: {"Ratio":"60%","Backbone":"InternVL3.5-38B"}. This record does not assert independent reproduction.

### FastV

Accuracy: 58.38. Result ID: 6a0c13f2fbc373592c01d117.

[Reporting paper](<https://www.sota2.com/research/paper/6a0bee84dbc836b87d82c339.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa74ae1e209131cea632b7.md>)

Recorded evaluation configuration: {"Ratio":"60%","Backbone":"InternVL3.5-38B"}. This record does not assert independent reproduction.

### InternVL2-2B

Accuracy: 58.36. Result ID: 6996d0735698665ec41c8cf3.

[Reporting paper](<https://www.sota2.com/research/paper/692fc434d6b0dd3342933465.md>)

Recorded evaluation configuration: {"Params":"2.2B"}. This record does not assert independent reproduction.

### VisionZip

Accuracy: 58.35. Result ID: 6a0c13f2fbc373592c01d119.

[Reporting paper](<https://www.sota2.com/research/paper/6a0bee84dbc836b87d82c339.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa760f1e209131cea6b33b.md>)

[Implementation](<https://github.com/dvlab-research/VisionZip>)

Recorded evaluation configuration: {"Ratio":"60%","Backbone":"InternVL3.5-38B"}. This record does not assert independent reproduction.

### Sequential (SFT)

Accuracy: 58.3. Result ID: 6a27ed5cfbc373592c257218.

[Reporting paper](<https://www.sota2.com/research/paper/6a279e5b90966ade919bfa4c.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7b1b1e209131cea86980.md>)

Recorded evaluation configuration: {"Size":"3B"}. This record does not assert independent reproduction.

### Majority voting@4

Accuracy: 58.1. Result ID: 6a27ed5cfbc373592c25721c.

[Reporting paper](<https://www.sota2.com/research/paper/6a279e5b90966ade919bfa4c.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6eb61e209131cea41b33.md>)

Recorded evaluation configuration: {"Size":"3B"}. This record does not assert independent reproduction.

### Visual Para-Thinker++

Accuracy: 58. Result ID: 6a27ed5cfbc373592c25726f.

[Reporting paper](<https://www.sota2.com/research/paper/6a279e5b90966ade919bfa4c.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a279e5b90966ade919bfa4c.md>)

Recorded evaluation configuration: {"Backbone":"Qwen2.5-VL-7B","Training Stage (MAC/RDMAO)":"MAC only"}. This record does not assert independent reproduction.

### F3A

Accuracy: 57.96. Result ID: 6a0c13f2fbc373592c01d11f.

[Reporting paper](<https://www.sota2.com/research/paper/6a0bee84dbc836b87d82c339.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a0bee84dbc836b87d82c339.md>)

Recorded evaluation configuration: {"Ratio":"40%","Backbone":"InternVL3.5-38B"}. This record does not assert independent reproduction.

### MobileNet-Qwen

Accuracy: 57.94. Result ID: 6996d0735698665ec41c8cf8.

[Reporting paper](<https://www.sota2.com/research/paper/692fc434d6b0dd3342933465.md>)

Recorded evaluation configuration: {"Params":"1.84B"}. This record does not assert independent reproduction.

### DivPrune

Accuracy: 57.78. Result ID: 6a0c13f2fbc373592c01d11d.

[Reporting paper](<https://www.sota2.com/research/paper/6a0bee84dbc836b87d82c339.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa77c41e209131cea7532a.md>)

Recorded evaluation configuration: {"Ratio":"40%","Backbone":"InternVL3.5-38B"}. This record does not assert independent reproduction.

### FastV

Accuracy: 57.31. Result ID: 6a0c13f2fbc373592c01d11c.

[Reporting paper](<https://www.sota2.com/research/paper/6a0bee84dbc836b87d82c339.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa74ae1e209131cea632b7.md>)

Recorded evaluation configuration: {"Ratio":"40%","Backbone":"InternVL3.5-38B"}. This record does not assert independent reproduction.

### Honeybee-Remake-SEED-200K

Accuracy: 56.8. Result ID: 6a0ac75247f34ce924f0b6df.

[Reporting paper](<https://www.sota2.com/research/paper/6a0a9cdedbc836b87d82b23f.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a0a9cdedbc836b87d82b23f.md>)

Recorded evaluation configuration: {"Model":"Qwen3-VL (4B)","Selection Ratio":"20%"}. This record does not assert independent reproduction.

### BASE

Accuracy: 56.7. Result ID: 6a0ac75247f34ce924f0b6dc.

[Reporting paper](<https://www.sota2.com/research/paper/6a0a9cdedbc836b87d82b23f.md>)

Recorded evaluation configuration: {"Model":"Qwen3-VL (4B)","Selection Ratio":"0%"}. This record does not assert independent reproduction.

Page 1 of 2. This is an index page, not the full collection.

- [Next index page](<https://www.sota2.com/research/sota/visual-hallucination-evaluation-on-hallusionbench.md?metric=Accuracy&page=2>): Follow this link to discover more pages.
