MiniMax releases its first financial report, with revenue soaring by 158%, and overseas revenue accounting for 70%

Source: Xinzhiyuan

Just now, MiniMax delivered its first post-IPO earnings report! Revenue surged by 158%, with 70% coming from overseas. MiniMax’s global footprint has been fully expanded. The AI Agent revolution is already underway with Chinese players in position.

On the eve of the Lantern Festival, MiniMax submitted its first annual performance report since going public to the Hong Kong Stock Exchange.

OpenAI is still burning money, but MiniMax directly faces market tests this time, showing an astonishing growth curve—

During the reporting period, the company achieved a total revenue of $79.038 million, a year-over-year increase of 158.9%, with 73% of revenue from international markets.

Gross profit reached $20.078 million, a 437.2% increase from the same period last year, with gross margin rising to 25.4%, significantly improving profitability.

At the same time, the company has served over 236 million users across more than 200 countries and regions, with 214,000 enterprise clients and developers.

Behind this growth curve, we finally see the scarce value of MiniMax.

First, MiniMax is transitioning from a large model company to a platform company in the AI era.

Founder and CEO Yan Junjie systematically explained: “AI platform value = intelligence density × Token throughput.”

He pointed out that the core of future AI industry competition lies in defining the boundaries of intelligence. When these boundaries are broken, new scenarios and users will emerge, further forming new ecosystems and commercial benefits.

Second, improvements in inference efficiency. The M2 series text models’ inference compute cost per million tokens has decreased by over 50% since December last year, and the Heluo video generation model’s inference latency has dropped by over 30% in the same period!

Additionally, MiniMax has built a comprehensive multimodal R&D capability, with models in language, video, speech, music, and other major modalities that are globally competitive.

Their global layout is also impressive, with MiniMax Agent, HailuoAI, Talkie/Xingye, and others having a large number of loyal users both domestically and abroad.

No wonder MiniMax’s results are so outstanding.

108 days to iterate 3 generations of models, fastest in the industry for programming progress

Why has MiniMax achieved such remarkable performance? Clearly, it is inseparable from their core technological advantages.

Over the past year, MiniMax’s models have been iterated at an ultra-fast pace, exemplifying the company’s technological dominance.

At the start of 2026, MiniMax launched the explosive product MaxClaw, directly dropping a heavy bomb into the market.

Today, MaxClaw is available on mobile.

It has been launched simultaneously worldwide on the MiniMax App (iOS & Android), allowing users to run OpenClaw directly on their phones for multi-device collaboration.

This “cloud AI claw” instantly disrupted the market’s calm. It is not a mild iteration but a naked revolution.

Based on the open-source OpenClaw framework, MaxClaw is equipped with the M2.5 model, deployable with one click, online in 10 seconds, truly an “intelligent agent accessible to everyone.”

Within just a few days, it spread wildly with rave reviews.

Amid the craze for “raising lobsters” in the AI circle, MiniMax seized the pain point, packaging the high-threshold open-source framework into a “zero-threshold revolution,” transforming developer toys into mass consumer products.

But the story is far from over; MaxClaw’s popularity is just the prelude to Agent upgrades.

Alongside MaxClaw’s launch, MiniMax Agent Expert 2.0 made a strong debut, allowing users to avoid the hassle of configuring Skills, SubAgents, MCP, and completely restructured AI agents!

MiniMax hasn’t stopped; they know that the foundation model is the core of intelligent agents. The popularity of MaxClaw is driven by the capabilities enhanced through model iteration.

Within 108 days, MiniMax updated three models: M2, M2.1, and M2-her, leading the industry in progress speed.

In the most representative programming benchmark SWE-Bench Verified, the M2 series models maintained the fastest improvement rate in the industry.

M2 also became the first Chinese model on OpenRouter with a daily token consumption exceeding 50 billion!

M2 redefined the balance of effectiveness, price, and speed, with key capabilities in programming, tool invocation, and deep search, leading the token consumption leaderboard.

Soon after, M2.1 was released, significantly improving performance on complex real-world tasks.

Additionally, the underlying models M2-her for Xingye and Talkie achieved the top global performance in 100-round long conversations.

The latest M2.5 model uses MoE architecture, with 229 billion parameters, activating 10 billion parameters per token, reaching inference speeds of 100 tokens/sec, with context windows from 200,000 to 1 million!

In coding, tool invocation, and logical reasoning, M2.5 shows clear advantages.

Behind the training of M2.5, the team designed an asynchronous native Agent RL system—Forge—that accelerates training by 40 times, breaking the “impossible triangle.”

In productivity scenarios like coding, tool invocation, search, and office work, it ranks among the top global players, comparable to Claude Opus 4.6, Gemini 3 Pro, GPT-5.2, and others.

Not only is performance strong, but cost-effectiveness is also high.

Running complex agents without cost constraints has been the original goal of the MiniMax team, and M2.5 is close to achieving this.

The version of M2.5 with 50 TPS costs only 5-10% of what Opus, Gemini 3 Pro, and GPT-5 models cost.

From the performance figures, where does MiniMax’s value lie?

Technological breakthroughs ultimately translate into business value.

As model capabilities improve and agent operating costs decrease, MiniMax’s commercial growth has a solid engine.

Based on the latest performance data, MiniMax’s “multimodal + integrated production + globalization” strategy is unleashing unique commercial value.

First, the company delivered a strong growth report: total revenue in 2026 reached $79.038 million, a 158.9% increase.

In revenue structure, two main engines drove this growth.

On one hand, AI-native product revenue reached $53.075 million, up 143.4%, proving the market appeal of consumer applications.

On the other hand, the open platform business for developers grew even faster, with revenue of $25.963 million, up 197.8%, demonstrating the powerful potential of B2B technology empowerment.

Efficiency in commercialization has also improved significantly. Thanks to major upgrades in model and system efficiency, as well as infrastructure optimization, gross margin increased from 12.2% in 2024 to 25.4%, a 13 percentage point improvement.

Meanwhile, adjusted net loss rate shrank significantly year-over-year, indicating that the company is moving toward a healthier financial model amid rapid expansion.

At the same time, user and ecosystem accumulation lay the foundation for sustained future growth.

As of now, MiniMax has served over 236 million users, with 214,000 enterprise clients and developers.

Especially noteworthy is that the number of new registered users on the open platform in February 2026 exceeded four times the level of December 2025, showing strong growth inertia.

Behind these impressive numbers are MiniMax’s multimodal capabilities, integrated production strategy, and global business ecosystem.

Multimodal: the efficiency behind cost

Currently, MiniMax has achieved remarkable results in multimodal capabilities.

Under an AI-native organizational structure, MiniMax has globally competitive models in language, video, speech, music, and other major modalities.

It is widely believed that achieving global leadership in multimodal AI requires at least a billion-dollar investment. MiniMax has done it with only about $500 million.

In February 2026, MiniMax released M2.5, capable of working continuously for an hour at 100 tokens/sec output speed, costing only $1.

This means that $10,000 can keep 4 agents running continuously for a year, making the operation of complex agents infinitely scalable economically!

Why insist on full multimodality?

From day one, MiniMax believed in this path’s feasibility.

Recent trends have also validated their judgment: in the past half-year, multimodal fusion has brought changes in intelligent boundaries, exemplified by Nano Banana Pro, which has gone viral online.

The industry trend is clear: future AGI will inevitably be multimodal models. Multimodal fusion is the essential path to enhance intelligence limits.

Today, this multimodal technological reserve has become a huge competitive advantage.

MiniMax is one of only three companies domestically that lead in every modality, and the only independent startup.

Integrated production as a moat

Integrated production has also become a major moat for MiniMax.

Currently, many companies tend to fall into two extremes: either only develop models or only develop applications. As a result, they lack either commercial monetization ability or technological barriers.

MiniMax, however, continues to invest in foundational models while scaling applications on the product side, laying out for sustained growth.

A set of data in the financial report strongly supports this: gross margin increased to 25.4%, up 13 percentage points from 12.2% in 2024.

The main reasons are improvements in model and system efficiency, as well as infrastructure optimization—

In MiniMax, the effects of scale are already evident.

In other words, the cost per unit of intelligence delivered has been pushed down.

From a cost perspective, the signals are even clearer.

First, sales and distribution expenses decreased by 40.3% year-over-year, from $86.995 million to $51.896 million—simple reason: good AI-native products mainly grow through organic growth and word-of-mouth, so promotional expenses naturally decline.

Second, R&D expenses increased by 33.8% (from $189 million to $253 million), far less than the 158.9% revenue growth, demonstrating significant efficiency improvements in R&D.

This aligns with the emphasis on R&D efficiency in the annual report.

Looking at the entire market, few companies can achieve all these: rapid revenue growth, rising gross profit, easier customer acquisition, and increased R&D investment simultaneously.

Overseas Market: Decisive for Ceiling

Additionally, MiniMax’s globalization strategy is well underway.

In 2025, revenue outside mainland China accounted for about 73%, truly establishing an overseas-led revenue structure.

Meanwhile, MiniMax’s business covers over 200 countries and regions, serving 236 million users, with 214,000 enterprise clients and developers from more than 100 countries.

In the face of giants like OpenAI, Anthropic, Google, and xAI “squeezing” the market, MiniMax has carved out a profitable path abroad.

Since launch, M2.5 has attracted significant international attention, with major cloud providers and AI-native cloud platforms deploying MiniMax models, including Google Vertex AI, Microsoft Azure AI Foundry, Fireworks AI, Nebius AI, and more.

On top coding platforms like OpenCode and Kilo Code, MiniMax has become the default choice for developers!

Recently, Notion also launched M2.5, its first and only open-source model.

This step is crucial for valuation potential.

Many domestic AI companies talk about “going overseas,” but most stay at “product localization”—selling the same product in new markets.

MiniMax, however, is genuinely being bought willingly and with real money by overseas users.

Clearly, MiniMax is truly moving toward an “AI platform company,” beginning to enjoy the business dividends of innovation and uniqueness.

Global AI Competition Enters the Second Stage

As MiniMax’s products are validated by users worldwide, it faces not just regional competition but steps onto the global AI stage.

Today’s global AI competition has entered the second phase.

From Claude Code’s explosive popularity to OpenClaw’s craze, intelligent agents are entering a golden era.

Claude Code’s success ended the era of handwritten code. OpenClaw ignited the spark of personalized agents.

When AI shifts from passive invocation to active work, the agent revolution truly erupts.

Y Combinator predicts a new economy driven by intelligent agents is forming.

Now, AI competition is entering the second stage: from coding to agents, with token volume and application scenarios set to further explode.

This trend is also evident in MiniMax’s financial report:

In February 2026, the average daily token consumption of the M2 series models increased over 6 times compared to December 2025, with token consumption from CodingPlan increasing over 10 times.

Within MiniMax, agent interns now cover nearly 90% of roles, including programming, data analysis, operations, HR, marketing, and sales.

MiniMax believes that in the next year, model intelligence will further improve.

They are betting on three major tracks: programming, office productivity, and multimodal creation:

In programming, AI will reach L4 to L5 levels of intelligence, moving from “tools” to “colleague-level” collaboration;

In office productivity, progress will mirror last year’s speed in programming;

In multimodal creation, content will become “directly deliverable,” with new forms like streaming and real-time output.

These three trends together mean a larger-scale explosion of intelligent supply and unprecedented innovation windows in applications.

MiniMax predicts that token consumption could increase by 10-100 times this year.

To meet these challenges, they have already laid out plans, with R&D on models M3 and Hailuo 3 series underway.

More importantly, MiniMax shared their internal thoughts on the future and bigger goals—

MiniMax: From Large Model Company to AI Platform Company

Strategically, MiniMax aims to evolve from a large model company to an AI platform company in the era.

Why does MiniMax have the chance to become a platform company?

First, AI is not a saturated market; annual incremental growth exceeds the existing stock, and there is no “winner-takes-all” scenario yet.

As long as there is innovation and uniqueness, there are opportunities.

MiniMax has already seized the initiative at the model, product, and ecosystem levels.

Next, they are preparing in two ways:

By breaking through the intelligent boundary with next-generation multimodal models to create model uniqueness;

By building unique products and ecosystems around these models.

Unlike big tech giants, MiniMax is one of the few companies capable of doing both, possibly the only startup in Asia.

They have this confidence—

In recent months, MiniMax’s model iteration speed, revenue growth, user service capacity, and token throughput have all been steadily advancing.

This allows them to more quickly define new model goals, as they fully leverage AI’s value internally, and each model upgrade significantly enhances their overall capabilities, including R&D and external services.

This creates a positive flywheel, which is the core competitive advantage of MiniMax.

Furthermore, they will deepen their commercialization and explore global market opportunities.

The impressive results in Hong Kong stocks are just the beginning.