Luo Fuli, head of Xiaomi’s large model team, said Artificial General Intelligence (AGI) could be realized within the next two years. In her view, the industry is entering a decisive stretch, and the coming months will be critical for testing both research depth and technical agility across major AI teams.
Chinese foundation models are catching up fast
Luo said Chinese companies, including Kimi and MiMo, have already developed foundation models with more than 1 trillion parameters. She argued that this progress has materially narrowed the gap with leading U.S. models in the pre-training stage, suggesting that domestic AI developers are advancing more quickly than before.
She added that the performance difference between leading Chinese models and top international systems has narrowed to roughly two to three months. Using frontier models such as Claude Opus 4.6 as a benchmark, Luo said the gap is no longer measured in years, but in much shorter development cycles.
AGI progress seen at 20% today
On the broader path toward AGI, Luo estimated that current progress stands at around 20%. If development continues at the present pace, she believes that figure could rise to 60% to 70% by the end of this year. The comment reflects a relatively optimistic outlook for how quickly model capabilities may improve over the next several months.
At the same time, her remarks highlight how important near-term execution has become. According to Luo, the next few months will not only shape advances in model performance, but also reveal which teams can move fastest in research, engineering, and technical adaptation.
Open-source Agent frameworks add momentum
Luo also pointed to the rise of open-source Agent frameworks as a major accelerator for AI research and development. She specifically mentioned OpenClaw, arguing that such frameworks amplify collective intelligence by allowing broader developer participation and enabling faster iteration.
From an industry perspective, expanding open-source ecosystems could strengthen the feedback loop between model capabilities, tooling, and deployment. Combined with larger Chinese foundation models, narrowing performance gaps, and more active open-source collaboration, the race toward AGI appears to be entering a more intense and fast-moving phase.