Baidu

Baidu AI Cloud unveiled a full-stack AI infrastructure for embodied intelligence at Baidu World 2026, built on its Baihe platform to cover Ego-centric video data processing, model training and deployment, cutting GPU time for pose recovery by 43%. Baidu says it already serves over 70 embodied AI companies and holds a 29.6% share of China's embodied AI cloud market.

At Baidu World 2026, Robin Li likened today's embodied AI to autonomous driving around 2016, suggesting the field still needs long-term technical buildup. He also framed autonomous driving as an indispensable part of AI competition and disclosed that Apollo Go, in partnership with Shenzhou Car Rental, runs the world's only robotaxi rental model, operating in Hainan for over a year.

Baidu unveiled its Tianchi 1024-card super node at the 2026 Baidu World conference, claiming a single unit can train 5-trillion-parameter models and scale horizontally to 10,000-card clusters, with launch planned for next year. The system uses optical engine interconnect for 3x card-to-card bandwidth, delivering 2.5x training and 3x inference performance gains, while a companion G3.5 KV Cache appliance doubles long-context inference throughput and cuts multi-turn TTFT by 90%. The timeline is roughly a year earlier than the 2028 launch announced last year.

At Baidu World 2026, Robin Li argued that rising token spend without matching revenue stems from swapping in AI tools without redesigning organization and workflows. He proposed three principles for AI-era org evolution and disclosed that Baidu's internal daily active agents grew from about 5,000 to 37,000, with weekly effective tasks up more than 11x.

At Baidu World 2026, Robin Li unveiled "Ciyuansuo," a neutral, trusted marketplace for buying and selling large model services. The platform runs admission tests and continuous quality monitoring on models and issues credit ratings for providers across service quality, compliance and commercial delivery, reporting over 100 supply resources and 700 billion daily tokens within its first month.

At Baidu World 2026, Robin Li disclosed that Baidu's internal daily active agents grew from 5,000 to 37,000 within a year, a more than 6x increase, while weekly completed effective tasks rose from 84,000 to over 1 million, an 11x jump. Baidu has also built 4,300 workspaces and accumulated over 20,000 organization-level skills for cross-team reuse and agent invocation.

At Baidu World 2026, Robin Li argued that the next act of AI is the "intelligent organization," describing an AI-era stack built on GPUs with cloud, models and agent applications layered above, while organizational capability downstream is where productivity is actually unlocked. He urged enterprises to move beyond treating AI as a personal productivity tool and embed digital employees into sales, support, R&D and operations.

Baidu unveiled ERNIE 5.5 Flash Preview, built on a new architecture optimized for efficient inference with 280B total parameters, 14B active per token, a 1M-token context window and over 200 tokens/sec generation. It reached 1668 points on the Code Arena WebDev leaderboard, edging past DeepSeek-V4.1-Flash and Kimi K3 and matching Qwen3.8-Max, with weights to be open-sourced under Apache 2.0 by November 25.

Baidu released ERNIE 5.5 Flash Preview, built on a new architecture optimized for efficient inference, with 280B total parameters and 14B active per token, a 1M-token context window, and generation speed above 200 tokens/sec. The model is now live on Arena for testing, with Baidu claiming competitive results on Terminal-Bench and AutomationBench; open weights are slated for release by November 25.

Baidu unveiled ERNIE 5.5 Flash Preview, a 280B-parameter MoE model activating only 14B per pass, with a 1M-token context and over 200 tokens/sec output at as little as ¥1 per million input tokens off-peak. It claims top scores on Terminal-Bench 4.0 and AutomationBench, demoed an Unreal Engine 5 wuxia game scene built in 11 hours, and will open-source its weights globally on November 25.