Skip to content

OpenCity

An open spatiotemporal foundation model for traffic prediction

Visit official site
Category AI models

What OpenCity is

The model is an open source spatiotemporal foundation model for traffic prediction that combines a Transformer and graph neural networks. It is pretrained on large heterogeneous datasets for zero-shot generalization across cities.

What you can do with it

  • Predict New York City traffic flow for signal control
  • Analyze taxi demand patterns in Los Angeles

Getting started

  • Clone or download the codebase and install Python dependencies
  • Download and prepare the pretraining datasets

Pros & cons

✓ What we like

  • Full code, data and model weights are provided
  • Traffic flow, taxi demand and bicycle trajectories are named

! What to watch out for

  • Users clone the codebase and set up a Python environment
  • A pretrained configuration must be chosen before running scripts

FAQ

What architecture is used?

The card names a Transformer architecture together with graph neural networks.

Is fine tuning needed?

The card describes zero-shot use across traffic prediction tasks without fine tuning.

Last reviewed: 2026-11-15

More LLM API platform tools

View all →

How we review