Researchers built an RL environment around a physics simulator of a flying Tesla Roadster with four air thrusters, gusty wind, and the reveal's two stunts: a ramp flip and a kickflip over a Cybertruck. An LLM flies it with one command every 0.2 s while the car keeps moving. The latest open models can't fly it in real time. Deep Seek V4.1 Flash, Nemotron 3.5 Super, Muse Spark Open and Beam 1.0 take 1 to 6 s per decision and land 0% of kickflips and 3% or less of ramp flips. GPT-6 Astra also lands 0% on the clock. RL post-training made a 4B model the best pilot. Qwen3-4B, trained for about $13, answers in 0.09 s and lands 90% of kickflips in real time. With the world paused for the model to think, the best open model, Deep Seek V4.1 Flash, lands 87% and the other open models land 58 to 62%. Even GPT-6 Astra lands only 74%.
Everything is open source, and you can fly your own model without a GPU. Storm kickflips, 384 held-out flights per model. Open circles are runs with unlimited time to think. In real time, Deep Seek, GPT-6 Astra, Muse, Nemotron and Beam land 0%, and the RL-trained 4B lands 90%.
Source: Hacker News · Summarized by HeadlinesBriefing