WeRide unveils WITT model, cutting AI token costs by 98%

WeRide Inc. on Thursday unveiled WITT (World Intelligence Toward Truth), a Physical AI Cognitive Foundation Model that introduces Atomic Physical Facts as the fundamental building blocks for machine understanding of the physical world, reducing token costs by as much as 98% compared with general-purpose AI models.

"WITT decomposes continuously evolving real-world environments into verifiable facts that can be identified, reasoned about and validated," the company said in a statement. The model, named after philosopher Ludwig Wittgenstein's proposition that "the world is the totality of facts," extracts three categories of Atomic Physical Facts from driving data: standard driving facts, multi-agent interaction facts and physically ambiguous conditions.

The model processes up to 10,000 minutes of vehicle-operation video per day on a single GPU, delivering 200 times greater data-processing efficiency in comparable workloads, according to WeRide. A single request generates more than 100 dynamic tags, enabling rapid retrieval and validation of driving footage. WITT achieves an average factual error rate approximately one-third that of leading general-purpose AI models in autonomous driving scenario understanding tasks, the company said, without naming the specific models used for comparison.

The launch addresses a fundamental challenge in Physical AI: vast amounts of real-world operational data contain noise, human interventions and inactive segments that limit their training value. WITT's four-stage pipeline — Fact Extraction, Fact Reasoning, Fact Verification and Fact Curation — converts every kilometer of driving data into structured learning signals. High-value long-tail scenarios are routed to WeRide's GENESIS simulation model for synthetic training, while high-frequency everyday scenarios support reinforcement learning workflows.

How WITT Changes the Economics of Autonomous Driving

WeRide's architecture pairs WITT with GENESIS, its proprietary general-purpose simulation model, creating what the company calls a Physical AI flywheel. WITT extracts and verifies physical facts from real-world data; GENESIS generates high-fidelity simulation environments and long-tail training scenarios based on those facts. Together, they train vehicle-side models that improve through both real-world experience and synthetic learning.

The efficiency gains are material. While general-purpose AI models often rely on hundreds of billions of parameters, WITT delivers strong performance with a significantly leaner architecture. The 98% reduction in token costs means WeRide can process its fleet's operational data — spanning 3,000 autonomous vehicles across 40 cities in 12 countries — at a fraction of the compute cost competitors face.

WeRide has obtained autonomous driving permits across eight markets including China, the UAE, Singapore, France, Switzerland, Saudi Arabia, Belgium and the US. Its Robotaxi services operate fully driverless in Guangzhou, Beijing, Abu Dhabi and Dubai. The company's L2++ ADAS solution, WRD 3.0, has won six consecutive China Urban Intelligent Driving Competition titles and been selected for nearly 30 vehicle programs, including models from Chery Exeed and GAC Aion.

Competitive Landscape and Investor Implications

The announcement positions WeRide against a growing field of autonomous driving and Physical AI players. Pony AI, a Guangzhou-based rival, has expanded its robotaxi presence to more than 20 cities globally and recently launched consumer-facing app access in Singapore through a partnership with ComfortDelGro. Uber, which partners with WeRide for robotaxi deployments in Europe and the Middle East, has pursued a partnership-driven strategy since selling its self-driving division in 2020.

Cathie Wood's ARK Investment Management added 22,000 shares of WeRide on July 14, according to public filings, alongside purchases of Pony AI and Kodiak AI, signaling conviction in the autonomous driving technology sector. WeRide trades on both the Nasdaq and Hong Kong Stock Exchange under the ticker WRD.

For investors, WITT's claim of one-third the error rate of general-purpose models — if independently verified — could give WeRide a structural cost advantage in scaling autonomous operations. The autonomous ride-hailing market is projected to grow from $400 million in 2023 to $45.7 billion by 2030, according to Markets and Markets, a compound annual growth rate of 91.8%. WeRide's ability to convert operational data into training signals more efficiently than rivals may determine which companies capture that growth profitably.

This article is for informational purposes only and does not constitute investment advice.