On July 18, 2026, Rockchip showcased its edge-side AI technologies and industry application achievements under the theme “AIoT 2.0: Defining Next-Generation Smart Hardware” at Booth H2-A325 of the World Artificial Intelligence Conference (WAIC) in Shanghai. The company demonstrated a dual-chip collaborative architecture based on its RK35xx-series main SoCs and the RK1828 edge AI co-processor. This solution has achieved deep native optimization for mainstream large models such as Gemma4 and Qwen3/Qwen3.5, enabling smooth on-device inference of models ranging from 2B to 7B parameters. Real-world testing shows that the first-token latency of the Gemma4 model is as low as 169.93 ms. Multiple mass-produced or near-mass-production end products and industry solutions were exhibited on-site, including: - **In-Vehicle AI BOX**: Built on the RK3576M + RK1828 architecture and powered by the Qwen3.5-Omni multimodal model, it enables local fusion processing of in-cabin voice, vision, and sensor data while safeguarding driving privacy. - **Smart Office Devices**: Integrated with Tencent’s WorkBuddy assistant, these devices support multi-role collaboration scenarios—from strategy formulation to deliverable output—and feature dedicated local compute interfaces for handling real-time and sensitive tasks. - **Yici Connected Mid-Sized Display Solution**: Based on the RK3588 + RK1828 dual-chip design and running the MeOS agent system, it supports offline multi-turn conversations, cross-brand IoT interoperability, edge-side communication processing, and local data anonymization. - **Hisense RGB-mini LED TV UX**: Leveraging the RK1828, this TV delivers real-time 2D-to-3D rendering without requiring native 3D content, representing a flagship consumer product for on-device large-model deployment. - **Robotics Portfolio**: Includes companion robots, humanoid robots, and PUDU’s CC1 PRO commercial cleaning robot. The latter integrates the RK1828 as a standalone AI accelerator, enabling offline stain recognition and path planning with a cleaning efficiency of up to 1,000㎡/h. - **Industrial Edge Computing Devices**: - *Glasnostic GBox* for factory quality inspection, supporting millimeter-level defect detection and natural language interaction; - *Wanwu Zongheng DA600 All-in-One System*, adaptable across multiple scenarios, supporting expansion via multiple RK182X modules and secondary development in Python/C++. Additionally, Rockchip unveiled its self-developed Clawchips edge-cloud协同 Agent solution, offering modular skill components—including ASR, image parsing, and device debugging—that accelerate AI product development through one-click Skills deployment. The company also revealed that its next-generation edge AI co-processor has completed core validation, achieving breakthroughs in multimodal large-model inference and high-concurrency agent scheduling, with an official launch planned for Q3 2026. Through its modular “main SoC + AI co-processor” design, combined with the RKNN3 toolchain, foundational model optimizations, and diverse real-world deployments, Rockchip has established a complete AIoT 2.0 commercialization pathway spanning chips, models, applications, and end devices.