Sage Box, New Member 2.0, and R-UniAD 2.0: SenseTime positions itself as the leading automotive AI supplier.
The AMW Read
Incremental product launch within a well-known segment and company; updates SenseTime's automotive portfolio but does not fundamentally alter the competitive landscape.
Sage Box, New Member 2.0, and R-UniAD 2.0: SenseTime positions itself as the leading automotive AI supplier.
At the 2026 Beijing International Automotive Exhibition, SenseTime (商汤科技) released a suite of 'full-scenario AI agent' products for integrated cockpit and autonomous driving under its SenseAuto brand. The lineup includes Sage Box, a three-layer in-vehicle AI brain combining a Sage edge model, Sage OS orchestration, and native agents; New Member 2.0, an upgraded in-cabin agent that progresses from conversational to task-execution capabilities; and R-UniAD 2.0, a generative autonomous driving system that supports L2 to L4 with world model and reinforcement learning. SenseTime also announced a Robotaxi collaboration with T3 Mobility, targeting a 2026 trial launch.
Why it matters: SenseTime's product blitz exemplifies the 'agent-first' pattern in automotive AI, where the vehicle becomes a multi-modal agent platform rather than a simple transport device. By integrating cockpit and driving into a unified agent architecture with edge inference, SenseTime is challenging incumbents like Horizon Robotics and Mobileye on cost and capability. The company's claim of 'zero token cost' via edge deployment — saving $30 per vehicle per day in cloud API fees — directly addresses the capital-compression pressure on automakers. This move also updates the competitive dynamics in the Chinese smart EV market, where integration depth and ecosystem partnerships (30+ OEMs, 188 models, 5.5M cumulative shipments) are becoming decisive moats.
Expert take: SenseTime is executing a classic 'vertical integration + edge economics' play that mirrors how hyperscalers used on-device AI to break cloud dependency. If its edge inference cost savings hold at scale, it could reshape procurement incentives for OEMs evaluating AI cockpit solutions. The R-UniAD 2.0's world model + RL training loop also signals a shift from imitation learning to self-play simulation, which could reduce data collection costs — a recurring pain point in autonomous driving. However, the open question remains whether the multi-agent orchestration layer (Sage OS) can achieve the reliability needed for safety-critical L4 deployment, especially in mixed-traffic Chinese cities.
