Driving Reasoning on Drive-Internal (test) benchmark leaderboard · SOTA2 Research

Driving Reasoning on Drive-Internal (test)

56.48First-Frame Joint AccuracyGPT-5

Evaluation Results

Method	Links
GPT-52025.07		56.48	47.97
GPT-52025.07		56.3	47.19
DriveAgent-R12025.07		51.34	45.42
Human2025.07		49.59	49.29
DriveAgent-R12025.07		45.27	43.29
Gemini-2.5-Flash2025.07		43.38	43.42
GPT-4.12025.07		43.18	43.43
Gemini-2.5-Flash2025.07		41.83	42.14
GPT-4.12025.07		39.99	42.14
Doubao-Seed-1.62025.07		38.98	41.37
Doubao-Seed-1.62025.07		34.92	40.22
Qwen2.5-VL-72B2025.07		32.97	39.61
Qwen2.5-VL-72B2025.07		32.76	38.8
Qwen2.5-VL-7B2025.07		27.58	29.84
Qwen2.5-VL-3B2025.07		24.06	24.98
Qwen2.5-VL-7B2025.07		24.01	24.98
Qwen2.5-VL-3B2025.07		23.64	22.63