# Multimodal Autonomous Driving Capability on DriveBench Source: [Multimodal Autonomous Driving Capability on DriveBench]() Collection index: [llms.txt]() Data updated: 2026-02-26T05:57:46.760Z Evaluation of autonomous driving capabilities on the DriveBench dataset, broken down by Perception, Reasoning, and Planning performance. 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 | Multimodal Autonomous Driving Capability | | Dataset (including any stated split) | DriveBench | | Metric | Perception Accuracy | | Unit | PERCENT | | Direction | Higher is better | | Results loaded | 17 of 17 | | 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 - [Perception Accuracy](): Higher is better; unit: PERCENT. - [Reasoning Accuracy](): Higher is better; unit: PERCENT. - [Planning Accuracy](): Higher is better; unit: PERCENT. - [Average Accuracy](): Higher is better; unit: PERCENT. ## Recorded results | Method | Perception Accuracy | Result date | Report paper | | --- | --- | --- | --- | | Qwen2.5-VL-3B-Instruct (Multi) | 45.63 | Not recorded | 697c3af8a8b01e76cc8da3f3 | | Qwen2.5-VL-3B-Instruct (Single) | 45.59 | Not recorded | 697c3af8a8b01e76cc8da3f3 | | GPT-5.1 | 45.56 | Not recorded | 697c3af8a8b01e76cc8da3f3 | | InternVL3-1B (Multi) | 43.5 | Not recorded | 697c3af8a8b01e76cc8da3f3 | | InternVL3-1B (Single) | 43.13 | Not recorded | 697c3af8a8b01e76cc8da3f3 | | InternVL3-78B | 42.01 | Not recorded | 697c3af8a8b01e76cc8da3f3 | | InternVL3-8B | 40.05 | Not recorded | 697c3af8a8b01e76cc8da3f3 | | InternVL3-14B | 39.83 | Not recorded | 697c3af8a8b01e76cc8da3f3 | | Qwen2.5-VL-32B-Instruct | 38.41 | Not recorded | 697c3af8a8b01e76cc8da3f3 | | InternVL3-2B | 37.71 | Not recorded | 697c3af8a8b01e76cc8da3f3 | | Qwen2.5-VL-7B-Instruct | 36.26 | Not recorded | 697c3af8a8b01e76cc8da3f3 | | Qwen2.5-VL-3B-Instruct | 35.46 | Not recorded | 697c3af8a8b01e76cc8da3f3 | | InternVL3-38B | 34.27 | Not recorded | 697c3af8a8b01e76cc8da3f3 | | InternVL3-1B | 33.26 | Not recorded | 697c3af8a8b01e76cc8da3f3 | | Llama-3.2-11B-Vision-Instruct | 31.59 | Not recorded | 697c3af8a8b01e76cc8da3f3 | | Llama-3.2-90B-Vision-Instruct | 27.26 | Not recorded | 697c3af8a8b01e76cc8da3f3 | | Qwen2.5-VL-72B-Instruct | 23.78 | Not recorded | 697c3af8a8b01e76cc8da3f3 | ### Qwen2.5-VL-3B-Instruct (Multi) Perception Accuracy: 45.63. Result ID: 69961aaa5698665ec45aaca2. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Memory (GB)":8.5,"Speed (tok/s)":28,"1st-token (s)":0.68,"Distillation":"Multi-Teacher Drive-KD"}. This record does not assert independent reproduction. ### Qwen2.5-VL-3B-Instruct (Single) Perception Accuracy: 45.59. Result ID: 69961aaa5698665ec45aaca0. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Memory (GB)":8.5,"Speed (tok/s)":28,"1st-token (s)":0.68,"Distillation":"Single-Teacher Drive-KD"}. This record does not assert independent reproduction. ### GPT-5.1 Perception Accuracy: 45.56. Result ID: 69961aaa5698665ec45aac92. [Reporting paper]() Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction. ### InternVL3-1B (Multi) Perception Accuracy: 43.5. Result ID: 69961aaa5698665ec45aaca1. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Memory (GB)":4.1,"Speed (tok/s)":45.7,"1st-token (s)":0.45,"Distillation":"Multi-Teacher Drive-KD"}. This record does not assert independent reproduction. ### InternVL3-1B (Single) Perception Accuracy: 43.13. Result ID: 69961aaa5698665ec45aac9f. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Memory (GB)":4.1,"Speed (tok/s)":45.7,"1st-token (s)":0.45,"Distillation":"Single-Teacher Drive-KD"}. This record does not assert independent reproduction. ### InternVL3-78B Perception Accuracy: 42.01. Result ID: 69961aaa5698665ec45aac98. [Reporting paper]() Recorded evaluation configuration: {"Memory (GB)":171.6,"Speed (tok/s)":4,"1st-token (s)":16.46}. This record does not assert independent reproduction. ### InternVL3-8B Perception Accuracy: 40.05. Result ID: 69961aaa5698665ec45aac95. [Reporting paper]() Recorded evaluation configuration: {"Memory (GB)":18.3,"Speed (tok/s)":32.6,"1st-token (s)":1.58}. This record does not assert independent reproduction. ### InternVL3-14B Perception Accuracy: 39.83. Result ID: 69961aaa5698665ec45aac96. [Reporting paper]() Recorded evaluation configuration: {"Memory (GB)":33.4,"Speed (tok/s)":17,"1st-token (s)":3.01}. This record does not assert independent reproduction. ### Qwen2.5-VL-32B-Instruct Perception Accuracy: 38.41. Result ID: 69961aaa5698665ec45aac9b. [Reporting paper]() Recorded evaluation configuration: {"Memory (GB)":69.5,"Speed (tok/s)":10.8,"1st-token (s)":2.36}. This record does not assert independent reproduction. ### InternVL3-2B Perception Accuracy: 37.71. Result ID: 69961aaa5698665ec45aac94. [Reporting paper]() Recorded evaluation configuration: {"Memory (GB)":6.3,"Speed (tok/s)":39.9,"1st-token (s)":0.67}. This record does not assert independent reproduction. ### Qwen2.5-VL-7B-Instruct Perception Accuracy: 36.26. Result ID: 69961aaa5698665ec45aac9a. [Reporting paper]() Recorded evaluation configuration: {"Memory (GB)":17.1,"Speed (tok/s)":32,"1st-token (s)":0.87}. This record does not assert independent reproduction. ### Qwen2.5-VL-3B-Instruct Perception Accuracy: 35.46. Result ID: 69961aaa5698665ec45aac99. [Reporting paper]() Recorded evaluation configuration: {"Memory (GB)":8.5,"Speed (tok/s)":28,"1st-token (s)":0.68}. This record does not assert independent reproduction. ### InternVL3-38B Perception Accuracy: 34.27. Result ID: 69961aaa5698665ec45aac97. [Reporting paper]() Recorded evaluation configuration: {"Memory (GB)":87,"Speed (tok/s)":7.3,"1st-token (s)":9.84}. This record does not assert independent reproduction. ### InternVL3-1B Perception Accuracy: 33.26. Result ID: 69961aaa5698665ec45aac93. [Reporting paper]() Recorded evaluation configuration: {"Memory (GB)":4.1,"Speed (tok/s)":45.7,"1st-token (s)":0.45}. This record does not assert independent reproduction. ### Llama-3.2-11B-Vision-Instruct Perception Accuracy: 31.59. Result ID: 69961aaa5698665ec45aac9d. [Reporting paper]() Recorded evaluation configuration: {"Memory (GB)":26.2,"Speed (tok/s)":16.8,"1st-token (s)":1.55}. This record does not assert independent reproduction. ### Llama-3.2-90B-Vision-Instruct Perception Accuracy: 27.26. Result ID: 69961aaa5698665ec45aac9e. [Reporting paper]() Recorded evaluation configuration: {"Memory (GB)":183.6,"Speed (tok/s)":2.7,"1st-token (s)":8.05}. This record does not assert independent reproduction. ### Qwen2.5-VL-72B-Instruct Perception Accuracy: 23.78. Result ID: 69961aaa5698665ec45aac9c. [Reporting paper]() Recorded evaluation configuration: {"Memory (GB)":146.5,"Speed (tok/s)":5.8,"1st-token (s)":4.26}. This record does not assert independent reproduction. Page 1 of 1. This is an index page, not the full collection.