China’s AGI-Next summit backs 2026 shift from chatbots to agents

The AGI‑Next summit write‑up is less about a single announcement and more about a clear shift in how China’s leading labs frame the race. Yang Zhilin (Moonshot/Kimi), Lin Junyang (Alibaba Qwen), Yao Shunyu (now at Tencent, formerly OpenAI) and Tang Jie (Zhipu/Tsinghua) all converge on a narrative: the era of “talking AIs” and pure parameter scaling is winding down; 2026 is about agents that actually do things in the world with far better “intelligence efficiency.” That includes coding and reasoning bets from Zhipu, world‑model and optimizer work at Moonshot, and Alibaba’s emphasis on long‑tail, embodied and enterprise agents.([huxiu.com](https://www.huxiu.com/article/4825438.html))

Strategically, this matters because it suggests Chinese labs are not simply chasing OpenAI on raw model size but are trying to differentiate on efficiency, agent architecture and integration with robotics and real‑world tasks. The frank acknowledgement of a compute gap with the U.S.—and Lin’s off‑the‑cuff estimate of only ~20% odds of Chinese global leadership in 3–5 years—also underscores how seriously they take that disadvantage. If they can turn resource constraints into an advantage by pioneering more efficient architectures and agent designs, that could feed back into the global AGI race in unexpected ways. At minimum, the summit signals that “agentic AI” is becoming the shared mental model for what comes after the chatbot wave, on both sides of the Pacific.