When the model begins to sprout platforms: MiniMax's turning point

```

As one of the most watched AI model companies in the Hong Kong stock market this year, MiniMax has delivered its first annual report card.

Alongside its financial report, it also showcased its ambition to transition from a "model company" to an "AI platform company." Against the backdrop of intensified global competition in large models and a capital market gradually shifting from "storytelling" to "results-focused," this financial report clearly serves as a bellwether.

Financial data shows that MiniMax achieved rapid revenue growth in 2025. Revenue from AI-native products soared from $21.805 million in 2024 to $53.075 million in 2025—a year-on-year increase of 143.4%. This was mainly attributed to the viral reputation and significantly increased willingness to pay for products such as Conch AI and Talkie/Xingye globally.

Meanwhile, revenue from open platforms and enterprise services targeted at the B-end also surged by 197.8% to $25.963 million, demonstrating the scale effect of its model API in enterprise-level applications. The change in revenue structure means MiniMax is breaking away from dependency on single model call revenues, gradually forming a dual-driven commercial pattern of C-end products and B-end services.

Against the backdrop of skyrocketing revenue, MiniMax’s gross profit leaped 437.2% to $20.079 million, with gross margin doubling from 12.2% to 25.4%.

According to industry insiders, MiniMax did not get caught up in the traditional Internet era’s old practice of "burning money for traffic," but successfully found a path for organic growth driven by product strength.

MiniMax’s COO Yun Yeyi told WallstreetCN that while domestic tech giants care most about DAU, MiniMax cares most about whether it can define the next generation of intelligence.

MiniMax’s product lead also said, "AI is an era where production determines demand. What should we look at? The simplest is to see which model users are willing to pay for."

Progress on the technical front is another main thread of this financial report.

In Q4 2025, MiniMax successively launched the M2, M2.1, and M2-her models, with the M2.5 version released in February 2026. Data show that the average daily token consumption of the M2 series text models in February 2026 was more than six times higher than in December 2025, with coding plan-related consumption increasing more than tenfold.

The explosive growth in token consumption means that the improvement in model capability is driving real demand expansion.

Company founder Yan Junjie proposed that in the coming year, model intelligence levels will further leap forward. The programming sector will enter L4 to L5 intelligence stages, transitioning from "tools" to "colleague-level" collaboration; the office scenario will mirror last year’s progress in programming; multimodal creation will achieve direct deliverable medium-to-long content, and even present new forms of real-time streaming output.

In his view, these three factors combined will bring a massive explosion of intelligent supply, opening up a new window for unprecedented innovation at the application layer. This judgment is both a forecast of technology trends and a declaration of corporate strategic direction.

For MiniMax, last year was a decisive year in crossing from single modality to full modal fusion.

Last year’s released video model Conch 2.3 significantly improved action performance and image quality, and reduced batch creation costs by up to 50%. By year-end, it had generated over 600 million videos; Speech model 2.6 supported more than 40 languages, enabled ultra-low latency interaction, and generated over 200 million hours of voice; Music model 2.5 achieved more complex emotional expression and longer output duration.

The synergy of language, video, voice, and music—the four major modalities—forms its "full-modal" technology matrix, providing the underlying support for products like Conch AI.

In addition to external product iteration, MiniMax also carried out an experiment at the organizational level. The company disclosed that its internal Agent interns have covered nearly 90% of employee positions, with AI collaboration introduced from programming development to sales and marketing. In early 2026, the company will productize its internal capabilities, launching the MiniMax Agent AI-native Workspace.

This "use internally first, then sell externally" path treats the company itself as an experiment field for AI-native organizations, aiming to boost product innovation by improving organizational efficiency. If this model works, MiniMax will not only be a model supplier, but could also become an infrastructure provider for enterprise-level AI productivity platforms.

Overall, this annual report delivers three key messages. First, the commercialization path is becoming clearer and revenue growth is sustainable. Second, gross margin has significantly improved—technical efficiency is now translating into financial efficiency. Third, model capabilities are rapidly advancing, expanding towards multimodality and agent directions.

Of course, challenges remain.

High R&D spending will continue to dampen profit performance; global computing power cost fluctuations and international competitive pressure cannot be ignored; the timeline for profitability is still unclear. In 2025, MiniMax’s R&D expenditure reached as high as $2.53 billion, up 33.8% year-on-year. Increased investment in cloud services and computing power training led to an adjusted net loss in 2025 still hitting $250 million.

In the capital market, as a star company in the Hong Kong AI sector, MiniMax’s valuation has already partly reflected expectations for high growth. Whether it can continue to deliver technological leadership and commercial expansion will become the core variable determining share price performance.

From a long-term perspective, MiniMax is at a critical inflection point. If L4 to L5 programming capabilities are truly realized in 2026, with office and multimodal content achieving large-scale commercial use, the transition from "large model developer" to "platform company" will have a real foundation. Then, what MiniMax provides will no longer just be model call demand, but could become the underpinning of a new generation productivity system.

The 2025 report card is a capability verification; the execution in 2026 will determine whether MiniMax can truly accomplish this strategic upgrade.

Risk Warning and DisclaimerThe market is risky, investment requires caution. This article does not constitute personal investment advice, nor does it take into account the special investment objectives, financial situation, or needs of individual users. Users should consider whether any opinions, viewpoints, or conclusions in this article are appropriate for their specific situation. Any investments made based on this are at your own risk. ```