MINIMAX-WP(00100) released its annual performance, with total revenue increasing by 158.9% year-on-year to reach $79 million.

MINIMAX-WP(00100) released its annual performance for the year ending December 31, 2025, with the group generating revenue of $79.038 million, a year-on-year increase of 158.9%; adjusted net loss of $2.51 billion, a year-on-year increase of 2.7%; and annual net loss of $18.72 billion, a year-on-year increase of 302.29%. In 2025, we established a full modal R&D capability, with language, video, voice, music, and other major modalities having globally competitive models. We continuously provide users worldwide with a better experience through technological innovation, enhancing our AI native products, including an open platform for enterprise customers, MiniMax Agent, Conch AI, Talkie/Hoshino, and more. Our global presence is expanding deeper and more solidly. In 2025, MiniMax's total revenue for the full year increased by 158.9% to reach $79 million, with over 70% of the revenue coming from international markets. As of December 31, 2025, MiniMax has served over 236 million users in over 200 countries and regions, as well as 214,000 enterprise customers and developers from over 100 countries and regions. In terms of language models, in the fourth quarter of 2025, we updated three models: M2, M2.1, and M2-her. M2 redefined the balance between performance, price, and speed, with key capabilities in programming, tool invocation, and deep search, performing close to the top global level in international authoritative tests. After its release, M2 quickly gained recognition from the global developer community, becoming the first Chinese model on OpenRouter to consume over 50 billion tokens daily and topping HuggingFace's global hot list. Building on M2, M2.1 focuses on improving performance in complex real-world tasks, especially in understanding and executing multi-step instructions in programming and office scenarios. The founder of the open-source project OpenClaw, Peter, found that M2.1 has higher accuracy in tool invocation and can efficiently handle full-stack tasks. Additionally, M2-her, as the underlying model supporting the AI interactive products Hoshino and Talkie, focuses on creating a more natural and personalized conversation experience, ranking first globally in a 100-round long conversation test. In February 2026, we released M2.5, which has globally leading capabilities in productivity scenarios such as programming, tool invocation, and office tasks. In programming, M2.5 set a new industry record in efficiency in SWE-Bench Verified tests, with a 37% efficiency improvement over the previous generation M2.1. Most importantly, M2.5 makes the operation of complex agents infinitely scalable in terms of economic costsan hour of continuous work at a speed of 100 tokens per second only costs $1, meaning $10,000 can keep four agents working continuously for a year. The breakthrough in model capabilities also drove a rapid increase in usage after its release, dominating OpenRouter. From M2 to M2.1 to M2.5, each generation of models has seen significant improvements in capabilities and usage. The average daily token consumption of the M2 series text models in February 2026 has increased by over 6 times compared to December 2025, with token consumption from the Coding Plan growing by over 10 times. Looking ahead, at the strategic level, we will transition from a big model company to a platform company in the era of AI, continually defining and driving new intelligent paradigms, enhancing the innovation capabilities of technology and products, as well as scalable infrastructure and token throughput capabilities. We will deepen our commercial layout and cultivate global market opportunities, providing more powerful intelligence for global users and partners.