Kimi K2.5 Drives Moonshot AI's Global Surge, Overseas Revenue Overtakes Domestic

The landscape of large language models (LLMs) is undergoing a significant shift, highlighted by the recent global breakout of Moonshot AI's Kimi. Following the launch of its advanced K2.5 model, the company has reached a critical inflection point: for the first time, Kimi's revenue generated from overseas markets has exceeded its domestic earnings. This marks a substantial victory for domestically developed AI in the international commercial arena.

Kimi K2.5 Fuels Explosive Global User Growth

The introduction of the K2.5 model catalyzed an immediate and dramatic surge in Kimi's global performance metrics. In the span of just a few days post-update, the platform demonstrated remarkable traction:

• Fourfold Increase in Paid Users: Global paid user subscriptions quadrupled shortly after the K2.5 release.
• Rising Popularity Index: On prominent aggregate platforms like Openrouter, Kimi rapidly ascended the ranks, positioning itself as the third most popular model, situated just behind industry leaders like Claude Sonnet 4.5 and Gemini 3 Flash.
• Rapid Revenue Escalation: Since November 2025, overseas API revenue has multiplied by four times, with both domestic and international monthly paid user growth rates exceeding 170% month-over-month.

This impressive financial and usage spike underscores a successful internationalization strategy, heavily supported by substantial advancements in core AI Technology.

Core Innovations: Native Multimodality and Agent Clusters

The underlying strength driving Kimi K2.5’s superior performance lies in fundamental architectural innovations. The model moved beyond iterative updates to adopt a comprehensively redesigned foundation:

Native Multimodal Architecture

K2.5 utilizes a native multimodal architecture, meaning its training incorporates various data types (text, vision, code) from the ground up, rather than layering modalities on top of a text-based foundation. This holistic approach ensures seamless and deep understanding across:

• Visual comprehension and analysis.
• Advanced code generation capabilities.
• Complex reasoning and cognitive patterns.

Agent Swarm Scheduling for Efficiency

A key differentiator introduced with K2.5 is the implementation of the “Agent Swarm” capability. This feature allows the model to effectively manage and coordinate multiple specialized agents in parallel to solve complex tasks. The system demonstrated the capacity to simultaneously dispatch up to 100 agents, processing approximately 1,500 sequential steps in complex information retrieval and synthesis tasks. This optimization resulted in efficiency gains ranging from three to ten times compared to previous methods.

Benchmark Performance

In rigorous testing environments, K2.5 achieved State-of-the-Art (SOTA) results on several critical LLM Benchmark suites, including HLE (Human-Level Evaluation) and BrowseComp. In some specific metrics, Kimi K2.5 even surpassed the performance indicators of established top-tier closed models, demonstrating genuine competitive parity on the global stage.

Product Evolution: From Chatbot to Productivity Engine

Moonshot AI’s leadership has clearly articulated a strategic pivot for Kimi. The focus is shifting decisively away from simply being a responsive chatbot toward becoming an indispensable C-side Productivity Tool. Founder Yang Zhilin emphasized that the team is expanding the model’s utility through the novel dimension of Agent Swarm coordination.

Unified Branding for Workflows

To align with this strategic direction, the in-testing product, previously known as OK Computer, has been formally rebranded as Kimi Agent. This change solidifies the product’s identity as a general-purpose office and workflow assistant.

Deep Editability and Complex Scenario Handling

Significant research and development efforts are now concentrated on deep editability within complex real-world scenarios. Examples of this focus include:

• Precise element splitting and editing within PowerPoint (PPT) presentations.
• Generating functional frontend code directly from screen recordings.
• Automated modeling and data structure creation within Excel sheets.

Strategic Focus Areas

The company has tightly defined its operational scope, concentrating resources on the foundational layers critical for high-value output. Moonshot AI is concentrating its expertise on the AI Agent layer, the core LLM foundation, and the logical reasoning layer. By maintaining a top-tier foundation model and deeply integrating sophisticated Agent capabilities, Kimi is successfully carving out a unique competitive niche essential for its Global Expansion.

Conclusion: A New Era for Kimi

The success of Kimi K2.5 illustrates the maturity of Moonshot AI’s technological roadmap. By prioritizing robust, natively multimodal capabilities and deploying advanced AI Model coordination via Agent Swarms, Kimi has transitioned from a strong domestic contender to a significant international player. The crossover of overseas revenue signals that the platform is effectively meeting the diverse and demanding needs of global users, setting a new precedent for future growth.