Hong Kong, Beijing and ShenzhenJune 9, 2026 /PRNewswire/ — Financial vertical AI large model company GIM (Grace Investment Machine) has recently completed its Angel+ round of financing, led by SAIF Partners, with participation from the family office of the CEO of an internet company valued at hundreds of billions of yuan. Previously, GIM’s Angel round was co-invested by Monolith Capital and 5Y Capital. To date, the company has completed over 100 million yuan in Angel and Angel+ round financing.
Founded in July 2025, GIM is dedicated to building an AI-native reasoning system specifically designed for investment research and decision-making. The company’s focus is not on simply transplanting general-purpose models into the financial domain, but on redefining the role of AI in research, judgment, and decision-making, starting from foundational capabilities.
Interdisciplinary Founding Team
GIM was co-founded by Xu Jiahao and Dr. Liu Qi. Before founding GIM, Xu Jiahao had long been deeply involved in both primary and secondary markets, having worked at 5Y Capital and Neumann Capital, and participated in investments in technology projects such as Horizon Robotics, XtalPi, and Pony.ai. In recent years, the leap in reasoning model capabilities has made him increasingly aware that AI’s role in investment will not remain at the level of an information tool but will gradually enter the decision-making process itself.
Founder Xu Jiahao speaking at the University of Hong Kong
Dr. Liu Qi is currently an Assistant Professor in the Department of Computer Science at the University of Hong Kong. His research focuses on large language models and multimodal AI. He earned his PhD from the University of Oxford under the supervision of Phil Blunsom and has conducted research at DeepMind and Meta FAIR.
GIM’s team also brings together interdisciplinary members from renowned hedge funds, quantitative funds, DeepMind, Meta, Microsoft, and other institutions. The company’s goal is clear: not to build general-purpose large models, but to create vertical reasoning infrastructure for the asset management industry.
$150 Billion Addressable Market, Opportunity Lies in General-Purpose Model Gaps
From a market size perspective, the combined revenue of the global financial AI software market and asset management services corresponds to an addressable space of approximately $150 billion. If financial large models truly enter the asset management system, they could unlock an even larger market worth $9 trillion.
But the greater opportunity lies not in the numbers themselves, but in the capability gaps of general-purpose models. General-purpose models can summarize financial reports, but they cannot truly answer the question, “Should I buy or not?” The issue is not a lack of financial knowledge, but a deficiency in the key capabilities needed for investment research: numerical reasoning, temporal awareness, compliance constraints, and dynamic updates. What the financial industry truly needs is never a system that merely repeats information, but a system that can organize research, form judgments, and continuously calibrate itself.
CogAlpha Multi-Agent Framework and Proprietary Financial Time-Series Large Model
In March 2026, a study published in the Journal of Accounting Research (JAR) pointed out that when users ask AI about stocks, over 70% directly request the model to tell them “what is worth buying.” As reasoning model capabilities leap forward, AI is transitioning from an information processing tool to a research tool—no longer just organizing materials and responding to commands, but actively deconstructing raw data and uncovering potential patterns.
GIM’s core exploration in this direction is the multi-agent framework CogAlpha. CogAlpha designs 21 specialized agents that form a complete AI investment research pipeline: some are responsible for assessing market risk states and cycles, others for analyzing the relationship between price and volume, and still others for identifying trends and reversal signals. Every newly proposed investment signal undergoes repeated review, questioning, modification, and evaluation within this pipeline; only if performance continues to improve after multiple rounds of assessment does it proceed to the next stage.
CogAlpha self-evolving signal mining process
Following this path, GIM further decided to develop its own proprietary financial vertical large model from scratch. Currently, this proprietary financial time-series large model has completed Scaling Law validation from 30M to 1.5B and then to 8B parameters. The model’s architecture introduces a time-series encoding mechanism and nonlinear gating structure tailored for financial data, and has demonstrated significant transfer learning capabilities in training across multiple frequencies, markets, and asset types.
Recently, CogAlpha’s research results have been accepted with high scores by the main conference of ACL 2026.