On June 1, MiniMax unveiled its new flagship model, M3.

This should have been a pivotal moment for MiniMax to solidify its position in the global technology landscape. M3 emphasizes long-context, multimodal, and code agent capabilities, with official claims that it approaches overseas leading closed-source models in various benchmark tests. The company aims to demonstrate that domestic large models still have room to break through with stronger performance and lower inference costs.

However, the spotlight from the technical launch was quickly overshadowed by simultaneous adjustments to billing rules.

After changes to the billing methods for MiniMax Code and Token Plan, some developers noticed significant changes in available quotas, call volumes, and user experience under existing plans. Many users questioned the platform's lack of adequate notice before the rule changes and raised dissatisfaction over plan benefits, token consumption rates, and inconsistent pricing across regions.

The model upgrade ultimately evolved into a crisis of trust.

This was not merely an operational mishap but rather a realistic challenge that large model companies must face after commercialization. When model upgrades bring higher inference costs, platforms must rebalance pricing, user experience, retention, and gross margins.

For an AI company that just went public with a valuation once pushed to lofty heights, any misstep in this balance can quickly be amplified by the capital markets into a revaluation of its business model.

Spotlight Fades: Billing Dispute Takes Center Stage

On the day of M3's release, MiniMax presented an aggressively technical narrative.

Officially, M3 supports a context window of up to 1M tokens, with significantly lower per-token computing costs in long-context scenarios and substantial improvements in prefill and decode speeds. In terms of code and agent capabilities, M3 also disclosed impressive test results, proving its ability to compete with global first-tier models.

However, for developers, improved model capabilities are only half the story—the other half is pricing and predictability.

Public reports indicate that the controversy initially centered on changes to billing access points. After M3's release, MiniMax shifted from a Coding Plan more focused on "per-call" usage to a billing system more reliant on token consumption. The low-cost entry point also shifted from the previous 29 RMB/month Starter tier to an M3 Token Plan starting at 49 RMB/month. What frustrated users even more was that multiple reports mentioned a lack of adequate notification via SMS or in-app messages before the adjustments, with many developers only discovering the rule changes upon logging in.

The perceived cost discrepancy soon amplified.

Some Plus-tier users reported on social media that whereas they could previously make around 1,500 calls within a 5-hour window, the revised plan only supported 300 to 500 calls in actual tests. Heavy users also calculated that, under equivalent monthly token consumption, actual payment costs could rise by as much as 257%. While these figures come from user-side estimates rather than official data, they sufficiently explain why a model upgrade quickly escalated into a crisis of trust.

Meanwhile, users also questioned the platform's pricing transparency regarding domestic and international API prices, different plan benefits, and weekly quota rules.

Price adjustments themselves are not inherently unreasonable; the issue lies in the fact that the core of developer tools is stable expectations. Once users feel that rules change too quickly, benefits are poorly explained, and notifications are inadequate, technological progress can easily be offset by commercialization anxieties.

MiniMax subsequently issued an apology and compensation plan, acknowledging shortcomings in user communication and handling of existing user benefits, and offered to preserve benefits, reset quotas, and provide additional quotas for some long-standing users.

However, remediation can hardly fully cover the initial loss of trust. For small and medium-sized development teams, APIs and development tools are not ordinary consumer products but infrastructure embedded in business processes. Once pricing becomes unpredictable, they will begin evaluating alternatives, regardless of migration costs.

This is the crux of MiniMax's controversy. While the outside world sees a pricing dispute, the capital markets see unverified commercialization capabilities.

Whether M3 can justify higher pricing depends on whether enterprise clients are willing to keep paying, whether developer call volumes remain stable, and whether retention rates hold after plan adjustments.

Ultimately, technical narratives must withstand scrutiny from commercialization data.

Stock Price Decline: More Than Just Public Sentiment

The stock price reacted swiftly.

On June 1, MiniMax's stock opened high but quickly plummeted, ending the day down 15.71%. The stock continued to face pressure afterward; by the Hong Kong stock market's close on June 12, MiniMax-W traded at 396 HKD, down over 70% from its March high of 1,330 HKD. The corresponding market value retracted sharply from its peak, evaporating nearly 300 billion HKD.

On the surface, this was backlash from the billing controversy. However, viewed over a longer timeline, MiniMax's stock adjustment was not merely a public sentiment issue.

The real backdrop was previously inflated valuations, a small free float, and strong premiums driven by the scarcity of AI assets.

Ultimately, the secondary market will not pay solely for imagination.

In 2025, MiniMax achieved total revenue of $79.038 million, up 158.9% year-over-year; gross margins improved from 12.2% in 2024 to 25.4%. Viewed solely through growth rates, this is an impressive performance. However, the absolute scale remains small, and losses persist.

In 2025, the company reported a net loss of $1.872 billion, mostly from fair value changes in financial instruments such as pre-IPO preferred shares, rather than operational cash losses. After adjusting for these effects, MiniMax's adjusted net loss was approximately $251 million, roughly flat compared to 2024.

This means the market's previous valuation of MiniMax was not based on current profitability but on high expectations for future platformization, globalization, and commercialization.

When M3's release did not enhance certainty but instead exposed friction between the billing system and user trust, valuations naturally began to loosen.

A deeper issue lies in MiniMax's revenue structure.

In 2025, MiniMax generated approximately $53.1 million in revenue from AI-native products, accounting for about 67.2% of total revenue; revenue from open platforms and other AI enterprise services reached approximately $26 million, or about 32.8%. The former includes C-end or C-end-oriented products like Talkie, Xingye, and Hailuo AI, while the latter covers APIs, open platforms, and enterprise services.

From a revenue contribution perspective, C-end remains MiniMax's foundation.

This is both an advantage and a pressure point. The advantage is that MiniMax is one of the few domestic companies to successfully launch global C-end AI applications, with products like Talkie and Hailuo AI demonstrating its ability to reach overseas users and form early commercial loops in scenarios like companionship, content, and video generation.

However, C-end AI-native products inherently face several challenges: low user migration costs, high volatility in content and emotional value consumption, the need for sustained investment in user acquisition and retention, and difficulty in quickly amortizing computing, moderation, and distribution costs.

In contrast, B-end APIs and enterprise services offer stronger gross margin profiles, higher client stickiness, and are more likely to be viewed by capital markets as sustainable business models.

MiniMax's challenge is that its B-end business volume remains insufficiently large.

In 2025, revenue from open platforms and other AI enterprise services grew 197.8% year-over-year, faster than AI-native products, but the base remained low. In other words, MiniMax has identified the direction for higher-quality revenue but has not yet completed the shift in its revenue structure.

With low-margin, highly volatile C-end business carrying the majority of revenue and high-margin, high-potential B-end business still in its growth phase, MiniMax's valuation can hardly rely solely on "leading model capabilities" until this structure changes significantly.

Lock-Up Expiration Looms: A Shift in AI Valuation Logic

An even more immediate pressure than short-term public sentiment is the upcoming lock-up expiration.

After MiniMax's Hong Kong IPO, its actual free float was small, amplifying scarcity premiums during the rally and liquidity pressures during the decline.

Starting in early July, the company will face a massive release of locked-up shares. Estimates from different brokerages on the proportion of shares to be unlocked vary, but CICC calculates that the unlocked shares will account for approximately 63% of the Hong Kong-listed share capital, with financial investors holding over one-third of this portion, while the current actual free float is only about 5%.

However, brokerages agree on one direction: the new supply of tradable shares will increase significantly.

For early investors, MiniMax still offers substantial unrealized gains despite the sharp correction. Given the combination of high volatility, elevated valuations, and unverified commercialization, some financial investors may reduce their stakes—unsurprisingly.

This is why the market is unlikely to see a trend reversal before the lock-up expiration actually occurs.

Of course, MiniMax is not without positive expectations. The company has been included in the Hang Seng Tech Index, and the market anticipates its potential inclusion in the Stock Connect. If southbound capital brings incremental allocation demand, it could indeed provide some liquidity support.

However, index funds and southbound capital act more as "supporting variables" than "revaluation variables." They can alleviate liquidity pressure but hardly resolve questions about the business model. Especially with a large lock-up expiration, whether new buying can absorb potential selling remains uncertain.

This is also key to MiniMax's current stock price pressure: the market no longer focuses solely on technical updates but begins calculating lock-ups, share reductions, gross margins, revenue structure, and cash burn.

Amid valuation pressure in the Hong Kong stock market, MiniMax has also initiated tutoring for a Sci-Tech Innovation Board listing, rapidly establishing an A+H dual-listing platform just 141 days after its Hong Kong IPO.

This sends a subtle signal: the company still needs broader capital channels to fund subsequent model research, computing investments, product expansion, and global competition. For large model companies, financing capability remains a core competitive advantage.

However, returning to A-shares is not a panacea.

Whether A-shares will grant MiniMax a higher valuation depends not on "being in the first tier of large models" but on how the market defines it. Is MiniMax a scarce full-modality foundational model platform or just another AI application and API supplier in a crowded field?

The former enjoys tech asset premiums; the latter will be re-evaluated within a more brutal framework of price wars and gross margins.

If MiniMax can demonstrate that M3 drives sustained B-end revenue growth, that its open platform becomes a second growth engine, and that C-end products contribute stable cash flow, A-share investors may indeed assign a more positive valuation.

However, if B-end transformation falls short, the API business gets dragged into industry price wars, and C-end products fail to significantly improve gross margins, A-share valuation premiums will not materialize out of thin air. Furthermore, if the stock price continues to decline, the value of equity incentives may shrink, potentially affecting the stability of core technical talent and creating a negative feedback loop.

The capital markets can give AI companies time, but not unlimited patience—and that patience is already waning.

In past years, AI companies could secure high valuations based on model parameters, leaderboard rankings, user scales, and technological visions. By 2026, the market has begun asking more concrete questions: Can revenue sustain? Can gross margins improve? Are clients willing to pay? Can token costs decline? Will there be capital to absorb shares after lock-ups? Can commercialization cover R&D investment?

MiniMax's stock price plunge is not in essence a rejection of AI but rather a refusal to believe in AI narratives unconditionally.

The M3 billing controversy was merely the trigger. It revealed that the stronger the model capabilities, the sharper the commercialization challenges. Platforms want higher revenue quality; users want lower costs and more stable benefits; companies want to sustain R&D investment; but the market demands a faster path to profitability.

For MiniMax, the key to success no longer lies in telling a grander story but in delivering harder operating data.

If the data delivers, the M3 controversy will be seen by the market as a temporary growing pain in commercialization. If the data falls short, MiniMax's valuation correction may only be the beginning.