# Vision-Language Perception on VL Perception Benchmarks Source: [Vision-Language Perception on VL Perception Benchmarks]() Collection index: [llms.txt]() Data updated: 2026-02-26T06:06:22.777Z Evaluation of model performance on the VL Perception Benchmarks dataset, covering Vision-Language Perception capabilities, measured by metrics including TextVQA and POPE scores. 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 | Vision-Language Perception | | Dataset (including any stated split) | VL Perception Benchmarks | | Metric | TextVQA | | Unit | PERCENT | | Direction | Higher is better | | Results loaded | 28 of 28 | | 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 - [TextVQA](): Higher is better; unit: PERCENT. - [POPE](): Higher is better; unit: PERCENT. - [Average Score](): Higher is better; unit: PERCENT. - [Seed](): Higher is better; unit: SCALAR. ## Recorded results | Method | TextVQA | Result date | Report paper | | --- | --- | --- | --- | | Base | 85.5 | Not recorded | 697c3aefa8b01e76cc8da38e | | Base | 85.5 | Not recorded | 697c3aefa8b01e76cc8da38e | | FRISM | 85.5 | Not recorded | 697c3aefa8b01e76cc8da38e | | DARE | 85.4 | Not recorded | 697c3aefa8b01e76cc8da38e | | FRISM | 85.4 | Not recorded | 697c3aefa8b01e76cc8da38e | | DARE | 85.4 | Not recorded | 697c3aefa8b01e76cc8da38e | | TA | 85 | Not recorded | 697c3aefa8b01e76cc8da38e | | IP-Merging | 84.8 | Not recorded | 697c3aefa8b01e76cc8da38e | | TA | 84.4 | Not recorded | 697c3aefa8b01e76cc8da38e | | TA | 84.1 | Not recorded | 697c3aefa8b01e76cc8da38e | | Ties-Merging | 83.8 | Not recorded | 697c3aefa8b01e76cc8da38e | | IP-Merging | 83.2 | Not recorded | 697c3aefa8b01e76cc8da38e | | TA | 82.9 | Not recorded | 697c3aefa8b01e76cc8da38e | | IP-Merging | 82.6 | Not recorded | 697c3aefa8b01e76cc8da38e | | IP-Merging | 82.5 | Not recorded | 697c3aefa8b01e76cc8da38e | | Base | 82.2 | Not recorded | 697c3aefa8b01e76cc8da38e | | FRISM | 82.2 | Not recorded | 697c3aefa8b01e76cc8da38e | | IP-Merging | 82.1 | Not recorded | 697c3aefa8b01e76cc8da38e | | DARE | 82 | Not recorded | 697c3aefa8b01e76cc8da38e | | TA | 81.8 | Not recorded | 697c3aefa8b01e76cc8da38e | | Ties-Merging | 80.3 | Not recorded | 697c3aefa8b01e76cc8da38e | | Ties-Merging | 74.3 | Not recorded | 697c3aefa8b01e76cc8da38e | | TA | 68.6 | Not recorded | 697c3aefa8b01e76cc8da38e | | TA | 62.8 | Not recorded | 697c3aefa8b01e76cc8da38e | | FRISM | 61.6 | Not recorded | 697c3aefa8b01e76cc8da38e | | IP-Merging | 61.3 | Not recorded | 697c3aefa8b01e76cc8da38e | | Base | 61.2 | Not recorded | 697c3aefa8b01e76cc8da38e | | TA | 61.2 | Not recorded | 697c3aefa8b01e76cc8da38e | ### Base TextVQA: 85.5. Result ID: 69977dfe5698665ec4d2862a. [Reporting paper]() Recorded evaluation configuration: {"Merged Models":"ThinkLite-VL-7B + DeepSeek-R1-distilled-Qwen-7B"}. This record does not assert independent reproduction. ### Base TextVQA: 85.5. Result ID: 69977dfe5698665ec4d28634. [Reporting paper]() Recorded evaluation configuration: {"Merged Models":"Qwen2.5-VL-7B-Instruct + OpenThinker3-7B"}. This record does not assert independent reproduction. ### FRISM TextVQA: 85.5. Result ID: 69977dfe5698665ec4d2863b. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Merged Models":"Qwen2.5-VL-7B-Instruct + OpenThinker3-7B"}. This record does not assert independent reproduction. ### DARE TextVQA: 85.4. Result ID: 69977dfe5698665ec4d2862f. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Merged Models":"ThinkLite-VL-7B + DeepSeek-R1-distilled-Qwen-7B"}. This record does not assert independent reproduction. ### FRISM TextVQA: 85.4. Result ID: 69977dfe5698665ec4d28633. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Merged Models":"ThinkLite-VL-7B + DeepSeek-R1-distilled-Qwen-7B"}. This record does not assert independent reproduction. ### DARE TextVQA: 85.4. Result ID: 69977dfe5698665ec4d28639. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Merged Models":"Qwen2.5-VL-7B-Instruct + OpenThinker3-7B"}. This record does not assert independent reproduction. ### TA TextVQA: 85. Result ID: 69977dfe5698665ec4d28635. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Merged Models":"Qwen2.5-VL-7B-Instruct + OpenThinker3-7B","lambda":0.1}. This record does not assert independent reproduction. ### IP-Merging TextVQA: 84.8. Result ID: 69977dfe5698665ec4d2863a. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Merged Models":"Qwen2.5-VL-7B-Instruct + OpenThinker3-7B"}. This record does not assert independent reproduction. ### TA TextVQA: 84.4. Result ID: 69977dfe5698665ec4d2862b. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Merged Models":"ThinkLite-VL-7B + DeepSeek-R1-distilled-Qwen-7B","lambda":0.1}. This record does not assert independent reproduction. ### TA TextVQA: 84.1. Result ID: 69977dfe5698665ec4d28636. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Merged Models":"Qwen2.5-VL-7B-Instruct + OpenThinker3-7B","lambda":0.15}. This record does not assert independent reproduction. ### Ties-Merging TextVQA: 83.8. Result ID: 69977dfe5698665ec4d28638. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Merged Models":"Qwen2.5-VL-7B-Instruct + OpenThinker3-7B"}. This record does not assert independent reproduction. ### IP-Merging TextVQA: 83.2. Result ID: 69977dfe5698665ec4d28632. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Merged Models":"ThinkLite-VL-7B + DeepSeek-R1-distilled-Qwen-7B","T":0.4}. This record does not assert independent reproduction. ### TA TextVQA: 82.9. Result ID: 69977dfe5698665ec4d28637. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Merged Models":"Qwen2.5-VL-7B-Instruct + OpenThinker3-7B","lambda":0.2}. This record does not assert independent reproduction. ### IP-Merging TextVQA: 82.6. Result ID: 69977dfe5698665ec4d28630. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Merged Models":"ThinkLite-VL-7B + DeepSeek-R1-distilled-Qwen-7B","T":0.2}. This record does not assert independent reproduction. ### IP-Merging TextVQA: 82.5. Result ID: 69977dfe5698665ec4d28631. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Merged Models":"ThinkLite-VL-7B + DeepSeek-R1-distilled-Qwen-7B","T":0.3}. This record does not assert independent reproduction. ### Base TextVQA: 82.2. Result ID: 69977dfe5698665ec4d2863c. [Reporting paper]() Recorded evaluation configuration: {"Merged Models":"InternVL3-8B + DeepSeek-R1-distilled-Qwen-7B"}. This record does not assert independent reproduction. ### FRISM TextVQA: 82.2. Result ID: 69977dfe5698665ec4d28641. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Merged Models":"InternVL3-8B + DeepSeek-R1-distilled-Qwen-7B"}. This record does not assert independent reproduction. ### IP-Merging TextVQA: 82.1. Result ID: 69977dfe5698665ec4d28640. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Merged Models":"InternVL3-8B + DeepSeek-R1-distilled-Qwen-7B"}. This record does not assert independent reproduction. ### DARE TextVQA: 82. Result ID: 69977dfe5698665ec4d2863f. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Merged Models":"InternVL3-8B + DeepSeek-R1-distilled-Qwen-7B"}. This record does not assert independent reproduction. ### TA TextVQA: 81.8. Result ID: 69977dfe5698665ec4d2863d. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Merged Models":"InternVL3-8B + DeepSeek-R1-distilled-Qwen-7B"}. This record does not assert independent reproduction. ### Ties-Merging TextVQA: 80.3. Result ID: 69977dfe5698665ec4d2863e. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Merged Models":"InternVL3-8B + DeepSeek-R1-distilled-Qwen-7B"}. This record does not assert independent reproduction. ### Ties-Merging TextVQA: 74.3. Result ID: 69977dfe5698665ec4d2862e. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Merged Models":"ThinkLite-VL-7B + DeepSeek-R1-distilled-Qwen-7B"}. This record does not assert independent reproduction. ### TA TextVQA: 68.6. Result ID: 69977dfe5698665ec4d2862c. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Merged Models":"ThinkLite-VL-7B + DeepSeek-R1-distilled-Qwen-7B","lambda":0.15}. This record does not assert independent reproduction. ### TA TextVQA: 62.8. Result ID: 69977dfe5698665ec4d2862d. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Merged Models":"ThinkLite-VL-7B + DeepSeek-R1-distilled-Qwen-7B","lambda":0.2}. This record does not assert independent reproduction. ### FRISM TextVQA: 61.6. Result ID: 69977dfe5698665ec4d28611. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Merging Setup":"LLAVA-Next-8B with DeepThought-8B"}. This record does not assert independent reproduction. ### IP-Merging TextVQA: 61.3. Result ID: 69977dfe5698665ec4d28610. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Merging Setup":"LLAVA-Next-8B with DeepThought-8B"}. This record does not assert independent reproduction. ### Base TextVQA: 61.2. Result ID: 69977dfe5698665ec4d2860e. [Reporting paper]() Recorded evaluation configuration: {"Base Model":"LLAVA-Next-8B"}. This record does not assert independent reproduction. ### TA TextVQA: 61.2. Result ID: 69977dfe5698665ec4d2860f. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Merging Setup":"LLAVA-Next-8B with DeepThought-8B","Method Type":"Task Arithmetic"}. This record does not assert independent reproduction. Page 1 of 1. This is an index page, not the full collection.