← XPENGTech

4 September 2026

VLA 6.3.0 Remembers 30 Seconds and Predicts 6 Seconds

XPeng's second-gen VLA adds the dimension of time: Infini-VLA's 30s memory and X-Foresight's 6s prediction — first on the G9L.

xpengvlag9laiautonomous-driving

AI's understanding of the world now includes the dimension of time — remembering the past, understanding the present, and predicting possible futures.

XPeng's Turing second-gen VLA 6.3.0 remembers the world 30 seconds ago and predicts what happens 6 seconds later — a step toward driving that surpasses humans.

  • Infini-VLA long-time-series architecture supports an effective time sequence of up to 30 seconds.
  • X-Foresight predictive world-model is on-vehicle, extrapolating the next 6 seconds.

The capability lands first on the XPeng G9L. The new VLA redefines space through time.

Figures

Fig 1

Fig 2

Fig 3

Fig 4


Source: 托马斯电火车 (Weibo) — translated & adapted for XPENGTech.

Get the next one direct.

New articles, or a daily / weekly digest — the same list EVConduit keeps.