23 Daily — According to leiphone.com, at the 2026 World Internet Conference in Wuzhen, Qianli Technology's CEO Chen Qi articulated that the core of intelligent driving competition lies in dual axes: the intelligence ceiling set by large-scale models and the efficiency of engineering delivery. By combining a native intelligent driving large model with a unique engineering system, Qianli advances both high-level R&D and mass production capabilities, promoting real-world applications of autonomous driving technology.

Transition from Rule-Based Models to Native Large Model Design

In recent years, the autonomous driving sector mainly relied on modular rules and iterative fixes on small models to extend scenario coverage. However, such approaches revealed limitations when deployed in real-world conditions. To effectively tackle complex long-tail scenarios and the rich variability of the physical world, higher demands have been placed on foundational models beyond patchwork and rule stacking.

Qianli Technology partnered with Jiyue Star to overcome constraints in funding and expertise, developing a native foundational large model for intelligent driving. Their training pipeline integrates driving data with general common sense knowledge during pre-training, enabling the model to inherently understand physical environments and driving contexts rather than retrofitting specialized capabilities via post-training fine-tuning.

Incorporating World Models and VLA for Enhanced Reasoning and Simulation

Qianli’s model continuously incorporates Visual-Language-Action (VLA) capabilities and world models to bolster its logical reasoning and real-world simulation. Chen noted that world modeling not only serves simulation but enables rational decision-making on real roads.

With scaling laws becoming more apparent, competition is increasingly favoring enterprises with robust large model development and deployment capabilities. Model quality and the extent of integration embody the true value in the upcoming stages of intelligent driving innovation.

Qianli’s 3-2-1 Engineering Framework Ensures Robust Mass Production

Qianli established a distinctive “3-2-1” R&D and engineering framework: three concurrent model development lines—one ensuring stable mass production, another driving technical implementation and production readiness, and a third exploring cutting-edge directions; two alternating AB versions support mass production and new model validation respectively, mitigating hardware and functional risks; one shared engineering base underpins data platforms and simulation evaluation, forming a unified data feedback loop between L2 mass-produced and L4 test vehicles.

This comprehensive system harmonizes continuous feature iteration with safety, cost-effectiveness, and replicability, propelling Qianli to capture nearly 20% market share among third-party city NOA providers in the first nine months of 2026. Their collaboration with Geely includes over 1 million deployed units of the Qianli Haohan system with an activation rate exceeding 92%.

Editor's take: Dual-Drive of Model Intelligence and Engineering Delivery as the Key to Autonomous Driving

Qianli’s progress exemplifies the industry’s shift from focusing on singular technological breakthroughs to balancing large model intelligence with engineering delivery efficiency. The model’s quality defines the intelligence ceiling, while the engineering system ensures stable and scalable mass deployment.

For entrepreneurs, this means autonomous driving solutions require building integrated ecosystems covering data, modeling, productization, and production pipelines—not just optimizing algorithms. Effectiveness can be verified through metrics such as model performance, data feedback velocity, iteration cadence, and user activation, aiming to strike a balance between technical advancement and commercial viability.