**[GPT-6 Astra Passed "Driving Test Part II"? 5 Minutes 22 Seconds to Park in Cone Slot]** Dynamic Observation Beating AI News: Three researchers from Taiwan, China, built DrivingBench, integrating general-purpose large models such as GPT-6 Astra, GPT-5.6 Sol, Grok 4.6, and Claude Fable 5.1 into a real 2022 Toyota Corolla to conduct real-world driving tests similar to "Driving Test Part II." The test track was approximately 130 meters long, consisting of an 8-meter-wide lane marked by colored cones. The vehicle had to navigate multiple turns and ultimately park in a 7×5 meter cone slot. GPT-6 Astra, which had never undergone specialized driving training, initially deviated from the track due to a judgment error during its first test. However, after adjusting its strategy for the second attempt, it completed the approximately 134.7-meter course in 5 minutes and 22 seconds and successfully parked in the slot, becoming the only model among the four to "pass Driving Test Part II." During the test, Astra controlled its speed near turns and obstacles at 0.5 to 0.8 meters per second and continuously adjusted its direction based on real-time camera footage. In contrast, Sol, Grok, and Fable all failed to complete the test. Sol and Grok failed three consecutive attempts, while Fable stopped during its third attempt after covering approximately 45% of the course. This test demonstrates that general-purpose large models without specialized driving data training can perform basic driving tasks in low-speed real-world environments, though there remains a significant gap compared to actual road autonomous driving. [Original Link]