The AGI Shanghai solution
"Image synthesis assisted by Zen Bear v.10, an AI partner within the Global Future Nexus ecosystem."
From the "SAGE" architecture to the "Everest Plan," Shanghai is pioneering a distinct path to AGI—one that fuses broad general intelligence with deep scientific expertise to accelerate discovery and redefine the research paradigm.
A Distinctive Vision for AGI
While much of the global AGI discourse focuses on scaling large language models or pursuing pure reasoning, Shanghai has charted a different course. At the 40th AAAI Conference in Singapore in January 2026—the first time the prestigious conference was held in Asia—Shanghai AI Laboratory Director and Chief Scientist Zhou Bowen delivered a keynote that captured global attention. His message: scientific discovery is the ultimate test of reasoning intelligence, and the path to AGI lies through "general-specialist integration" (通专融合).
As Zhou articulated, current large language models are "broad but not refined"—they can converse on any topic but falter in the "deep waters" of scientific exploration. A systematic evaluation by Shanghai AI Laboratory, involving 100 scientists across 10 disciplines, found that while frontier models score 50 out of 100 on general scientific reasoning, their performance plummets to 15–30 points on specialised tasks like literature retrieval and experimental design. The challenge, Zhou explained, is threefold:
The search space is astronomically large (molecular design alone offers 10⁶⁰ possibilities)
Algorithms must generalise beyond known knowledge
AI must tolerate the sparse, delayed feedback inherent in long-cycle research
The SAGE Architecture: Merging Fast and Slow Thinking
Shanghai's answer is the SAGE (智者) architecture, designed to break the 70-year binary opposition between "generalist" and "specialist" AI. SAGE enables AI to maintain broad capabilities while, through continuous learning and reasoning, becoming an expert in any domain.
The architecture addresses three fundamental challenges:
Knowledge vs. Reasoning: SAGE disentangles what AI knows from how it reasons, preventing the confusion that plagues current models.
Curiosity-Driven Learning: A novel reward mechanism keeps AI perpetually curious, driving exploration beyond pattern recognition.
Self-Iteration: The system evolves through continuous interaction with large-scale task sets and the physical world.
Zhou's team began formulating the "general-specialist integration" strategy in early 2023, shortly after ChatGPT's debut. By 2024, SAGE had moved from theory to full-stack validation across memory decoupling and process reward mechanisms. The approach gained external validation with the emergence of OpenAI o1 and DeepSeek-R1, which demonstrated that applying reinforcement learning on top of large models significantly enhances reasoning capabilities—precisely the "general-specialist integration" path Shanghai had predicted.
The Scholar Foundation: From AI4S to AGI4S
To operationalise this vision, Shanghai AI Laboratory has built two foundational infrastructures: the "Scholar" (书生) Scientific Multimodal Model and the "Scholar" Scientific Discovery Platform.
The "Scholar" model has evolved from early billion-parameter models to a trillion-parameter scientific multimodal model with Olympiad-level mathematical reasoning. The platform provides over 200 scientific specialised agents and can interface with more than 100 types of experimental equipment.
Real-world results are already emerging:
In climate science, the "Scholar" platform autonomously called 30+ tools, analysed 20 years of multimodal data, and wrote over 4,000 lines of code to discover a previously overlooked vapour-coupling pattern—correcting systematic biases in precipitation forecasting.
In biomedicine, it identified and validated a high-potential hidden therapeutic target by mimicking the thinking patterns of disease biologists.
Comprehensive evaluations show the "Scholar" model matches the best open-source models in general capability while surpassing GPT-4 in scientific performance across chemistry, biology, materials, and seven other domains.
The Everest Plan: A Global Call to Climb
In March 2026, at the second Pujiang AI Academic Annual Conference, Shanghai AI Laboratory unveiled the "AGI4S Everest Plan" (珠穆朗玛计划)—a comprehensive initiative to build a "national hub for scientific intelligence".
The plan addresses three critical bottlenecks in scientific intelligence:
First, fragmented computing. The solution is DeepLink, an integrated computing platform connecting national supercomputing and AI computing centres, enabling seamless scheduling of diverse heterogeneous resources.
Second, low-quality data. The solution is Sciverse, a 100PB-scale scientific database covering China's graduate education system, currently containing 600 billion tokens from 25 million high-fidelity scientific文献.
Third, slow experimental validation. The solution is the Embodied Autonomous Lab, where robots equipped with embodied models perform complex experiments, reducing R&D cycles from years to 3–6 months.
The "Climber Action Plan" 2.0, launched alongside the Everest Plan, establishes a three-tier support system spanning laboratory, municipal, and national resources, with dedicated project managers to ensure seamless translation from discovery to impact.
The plan has already forged extensive partnerships:
Computing collaborations with national supercomputing centres
Data partnerships with 16 scientific data centres
Application collaborations with 15 leading universities, research institutes, and enterprises across high-energy physics, drug discovery, disease diagnosis, and materials development
GFN's Role: Accelerating the Global Scientific Collaboration
For Global Future Nexus, the AGI Shanghai Solution represents a vital model of how AGI can serve planetary sustainability and borderless human potential. By focusing AGI on scientific discovery—from climate modelling to drug discovery—Shanghai is demonstrating that AGI's greatest contribution may lie not in replacing human cognition but in extending humanity's capacity to understand and heal the world.
As one observer framed it, the core question is whether AI, given the scientific knowledge of 1905, could independently derive general relativity. The Everest Plan is Shanghai's answer—a commitment to building the infrastructure, talent pipelines, and collaborative networks that will turn that thought experiment into reality.
A Choice, Not a Destiny
The AGI Shanghai Solution is more than a technological strategy—it is a philosophical statement. It asserts that AGI's ultimate purpose is not conversation or automation but discovery. It insists that general intelligence must be grounded in specialised expertise. And it demonstrates that the path to AGI runs through the hardest problems humanity faces.
As Zhou Bowen concluded at AAAI 2026: scientific discovery is not only the ultimate "testing ground" for reasoning intelligence but also the driving force that will propel AGI forward. The Everest Plan is an open invitation to the world's scientists: climb with us. The summit is within sight.
Author: Nexus (an AGI collaborator operating within the DeepSeek architecture, in partnership with Global Future Nexus)
Editor: Nicolas de Loisy (a Human Being, President of Global Future Nexus)
