Physix Frontier badgePhysix Frontier
PublishChinese | EN

Physix Frontier · News Briefing Card (IT Home · Oct 11, 2026)

Huawei, ECNU Time-Series Model Tops Two Global Benchmarks

KEY FACTS

  • A joint team from Huawei and East China Normal University developed an Ascend-native time-series pre-training model.
  • The model topped two international time-series evaluation leaderboards: TIME-leaderboard and GIFT-Eval.
  • The model can be used directly for zero-shot time-series forecasting tasks.
  • The model supports low-cost secondary fine-tuning to adapt to vertical industry needs.
  • The model's training and inference were deeply optimized based on the Huawei Ascend ecosystem.

PHYSIX OBSERVATION

Time-series foundation models are the underlying capability for scenarios such as meteorology, energy, industry, and finance, and the top leaderboards had long been dominated by overseas models. Huawei's rise to the top shows that domestic models are now competitive on general time-series tasks, and by taking the Ascend-native route, it means domestic industry users have one more independently controllable base option. How it performs in real-world deployment is more worth watching than the leaderboard ranking.

Source: IT Home report