Xiaomi has revealed a new artificial intelligence (AI) model named MiMo-V2-Pro, which has nearly one trillion parameters and can effectively compete with leading AI systems in the United States while operating at a considerably lower cost.
According to reports, this model achieves performance metrics that are comparable to those of OpenAI and Anthropic, requiring significantly fewer resources for each interaction.
Additionally, it is engineered to handle up to 256,000 tokens per session.
Under the leadership of a seasoned expert from the groundbreaking DeepSeek R1 initiative, Fuli Luo, this launch has been described as a “quiet ambush” on the global stage.
In a post on X, Fuli Luo said that the company intends to open source a variant of this model from the latest release, “We will open-source when the models are stable enough to deserve it.”
It is important to note that the new AI signifies the highest documented performance for a model of Chinese origin within this category.
The independent benchmarking organisation Artificial Analysis has verified the claims, ranking MiMo-V2-Pro at #10 on its global Intelligence Index with a score of 49.
This positions it alongside GPT-5.2 Codex and ahead of Grok 4.20 Beta.
Did Xiaomi develop an AI more advanced than ChatGPT and Gemini?
These findings further indicate that Xiaomi has effectively developed a model capable of the advanced reasoning necessary for engineering and production tasks.
MiMo V2 Pro is compatible with five primary frameworks for developing AI agents, including OpenClaw.
As stated in Xiaomi’s press release, the new flagship base model is designed for real-world agent workloads.
It functions as the “brain” within agent systems, enabling it to coordinate intricate workflows, oversee production engineering tasks, and deliver high-performance outcomes.
The algorithm surpassed Claude 4.6 Sonnet in the domain of program code generation, and its overall performance in agent tasks is nearly on par with that of Opus 4.6.
The optimisation of the algorithm’s training process has also led to enhanced performance stability and accuracy in tool calls, achieved through improvements in the learning process.
Read more: Xiaomi budget and flagship phones comparison which one is worth it
