MiniMax stands at a crossroads: lifting restrictions, price hikes, and triple strangulation of global giants

Recently, MiniMax released a new generation flagship model M3, which will switch from the long used per use billing to token billing on the same day. The 29 yuan monthly package will be directly increased to 49 yuan without prior notice to users.

The developer community exploded instantly, with token consumption far exceeding expectations for the same task. Some users estimated that the actual cost increase was as high as 257%.

Overnight, MiniMax transformed from a tech star to a backstabbing consumer.

Why did an AI unicorn with a market value of over 300 billion Hong Kong dollars fall from grace with just one price adjustment?

The price hike controversy is not an isolated operational accident, but a concentrated outbreak of deep-seated contradictions in MiniMax’s business model: relying on C-end virtual lovers to make quick money and support the development of expensive base models.

This path is becoming increasingly difficult to take in 2026, as computing power costs continue to rise and price wars intensify.

1、 The cost of cutting first and reporting later
Before the release of M3, MiniMax had a pretty good reputation in the developer community.

Its Coding Plan adopts a “per use deduction, no weekly limit” model, which is almost unique in the domestic AI programming service market.

In addition to its aggressive pricing strategy, the Starter price was only 29 yuan/month (9.9 yuan in the first month), far lower than the mainstream manufacturers’ price range of 40 to 50 yuan. MiniMax quickly accumulated a large number of individual developer users.

The 2025 annual report shows that the Token consumption of Coding Plan has grown rapidly, becoming a key engine for revenue growth on open platforms.

However, the release of M3 completely changed the rules of the game.

The new model adopts billing based on token consumption, and the minimum on sale package has jumped from 29 yuan to 49 yuan. What is even more difficult for users to accept is that there was no prior notice for this adjustment.

Many developers only realize that the rules have changed when they log in, and the monthly quota is exhausted within a few days.

A user who purchased the Plus package reported on social media that they were able to make about 1500 calls within the previous 5-hour window, but after the change, they were only able to support 300 to 500 calls. Another user calculated that in the past, consuming 3 to 5 billion tokens per month only cost 49 yuan, but now the same amount would cost about 175 yuan.

Adding fuel to the fire, MiniMax has been accused of setting two sets of standards for API pricing both domestically and internationally.

Taking context less than 512K as an example, the domestic input price is 2.1 yuan/million tokens, while the foreign input price is 0.3 US dollars (equivalent to about 2.04 yuan). Although the difference is not significant, at a point where user emotions are already tense, this differential treatment is quickly amplified into a crisis of trust.

In the face of overwhelming complaints, MiniMax’s parent company, Xiyu Technology, urgently issued an apology notice on the evening of June 1st, admitting that “this adjustment was not fully communicated with everyone in advance” and “the handling of issues such as the weekly limit for old users was not appropriate enough”.

The company subsequently introduced compensation measures: retaining the unlimited weekly subscription rights for old users who subscribed before March 22, giving new users an additional 50% quota, and uniformly resetting the usage quota from June 1 to 7.

From a technical logic perspective, this price increase is inevitable. The computing power consumption of the M3 model far exceeds that of the previous generation, and the original pay per use model can no longer withstand it in the era of intelligent agents.

Anthropic, OpenAI and other global peers also adopt token based billing, which is not a problem in itself. The problem with MiniMax lies in its execution: it does not provide users with sufficient buffer period, and does not explain the magnitude and reasons for price increases in advance. In a highly price sensitive developer market, this approach of cutting first and reporting later is tantamount to self destructing the Great Wall.

The underlying issue is that this price increase has exposed the fragility of MiniMax’s business model.

More than 70% of the company’s revenue comes from overseas C-end products (mainly the virtual companion application Talkie), while the gross profit margin of the C-end business is only 4.7%. In the technology industry, this is almost doing charity.

When the cost of computing power continues to rise, the original low price strategy is destined to be unsustainable, and price increases are an inevitable choice.

But how to raise prices and communicate with users tests the basic skills of a company in transitioning from technology driven to platform driven. From this perspective, MiniMax still has a long way to go.

2、 Standing awkwardly in the price war
If we only look at technical indicators, the performance of the M3 model is actually quite impressive.

On SWE Bench Pro, M3 scored 59.0%, surpassing GPT-5.5 (58.6%) and Gemini 3.1 Pro (54.2%), only behind Claude Opus 4.7. Scored 83.5 on BrowseComp, even surpassing Opus 4.7’s 79.3.

More importantly, M3 is the world’s first open-source model that integrates the three elements of “cutting-edge programming capabilities+1M ultra long context+native multimodality”. Vercel CEO Guillermo Rauch gave a review: “The MiniMax M3… is only slightly inferior to the Opus&GPT-5, but it is 10 times cheaper. ”

Today, as Silicon Valley giants move towards closed source, MiniMax insists on open-source the most advanced large-scale models, which has earned it a good reputation in the global developer community.

OpenRouter data shows that after the release of M3, the token consumption quickly exceeded 500 billion, and the M2 series once ranked among the top ten in global call volume rankings.

The co-founder and COO of the company, Yan Yeyi, disclosed at the end of May that the global enterprise and developer customer base has exceeded one million, a fivefold increase from six months ago, and ARR has doubled in the past two months.

But the problem is that the capital market has switched valuation logic. In the first half of this year, AI shifted from infrastructure frenzy to application monetization, and investors are no longer satisfied with benchmark scores. What they want is verifiable commercial data.

The pricing strategy of M3 just hit the muzzle of the price war.

Compared horizontally, M3’s situation is quite awkward. Compared with overseas giants, it does have the cost-effectiveness advantage of “domestic substitution”: in Claude Opus 4.7 standard mode, each million input tokens costs about $5 (about 35 yuan) and outputs $25 (about 175 yuan), which is several times that of M3.

But in the domestic battlefield, the pricing of M3 is significantly higher than the recently announced price reductions of DeepSeek V4 series and Xiaomi MiMo-V2.5 series. DeepSeeker V4 Flash costs only $0.14 (approximately 1 yuan) per million input tokens and $0.28 (approximately 2 yuan) per output. Even though the original price of Alibaba Qwen-3.7 Max is higher, it is still cheaper than M3 after a long-term 50% discount.

This means that M3’s “expensive for a reason” only holds true in specific high-end scenarios (such as long context, complex programming); In the cost sensitive mass B-end market, its cost-effectiveness advantage is not obvious.

This is precisely the fundamental reason why MiniMax quickly plunged after opening high on the day of M3 release (June 1st) and ultimately closed down 15.71%. The market voted with their feet to express doubts about whether the high pricing strategy could work.

It is worth noting that MiniMax is not unaware of the urgency of B-end transformation. The management of the company has clearly stated that the number of enterprise users is rapidly increasing and is already split in half with the revenue from the C-end.

But the problem lies in the fact that there is still a serious test in between, from the half to the B-end becoming a true growth engine: can M3 achieve both quantity and price increases in real enterprise scenarios?

The upcoming mid year report will be the first touchstone. If the growth rate of B-end revenue slows down and the gross profit margin does not improve, the narrative of successful switching to B-end will be falsified, and the stock price will be difficult to bottom out.

3、 July Siege: Countdown to Unlocking Flood Peaks
If the price hike controversy is an internal concern, then the upcoming lifting of the ban on Hongfeng and the listing of global AI giants are external threats.

Let’s first look at the pressure of lifting the ban. According to analysis by China International Capital Corporation, MiniMax will face a large-scale lifting of restricted shares on July 9th, with the unlocked shares accounting for about 63% of the Hong Kong stock capital, of which financial investors hold more than one-third.

At present, the actual circulation of the company’s inventory is only about 5%, and the supply will skyrocket by nearly 10 times after the lifting of the ban. Early investors had a substantial floating profit relative to the issue price (HKD 165), with some institutions having even higher floating profits.

In the current fragile market sentiment, selling pressure is almost inevitable. HSBC’s estimate is more direct: currently, MiniMax’s monthly cash consumption is about $28.1 million, and the support for the stock price after the lifting of the ban can only rely on whether the fundamentals can provide a better than expected answer.

At the same time, global AI giants are intensively entering the capital market.

Anthropic has completed a $65 billion H-round financing with a valuation of $965 billion and has secretly submitted an IPO application; OpenAI has also initiated the listing process; SpaceX was listed on NASDAQ on June 12th.

When these top global AI assets enter the public trading market, the valuation coordinates of Hong Kong stock big model companies will undergo a fundamental shift.

Taking Anthropic as a reference: Its latest ARR is about $47 billion, corresponding to a P/ARR of about 20 times, and it is expected to achieve profitability in the second quarter of 2026.

The current market value of MiniMax is about $27.4 billion, with an ARR of about $300 million and a P/ARR of about 60 times. Even if MiniMax can achieve an ARR of $1 billion by the end of the year (management goal), the P/ARR will still be about 19 times, basically aligned with Anthropic.

This means that a significant portion of the current valuation of nearly 60 times is based on market expectations of high ARR growth and the scarcity premium of Hong Kong stocks. Once scarcity disappears, valuation will face a systematic pullback.

Under such pressure, MiniMax signed a Sci Tech Innovation Board IPO guidance agreement with CITIC Securities on May 29th, launching the “A+H” dual capital platform layout. This move is both a defense and a gamble.

From a defensive perspective, the market value of Hong Kong stocks has significantly decreased and their refinancing capabilities have been compromised. The Science and Technology Innovation Board can provide lower financing costs and a more stable valuation anchor.

Historical data shows that A-shares typically have an overall premium of around 20% over H-shares, and the premium under the hard tech label may be even higher.

This funding can replenish key ammunition for MiniMax before the turning point of profitability arrives. In 2025, the company’s research and development expenses will be $253 million, and the net outflow of operating cash flow will be $280 million, indicating a tight funding chain. The net loss for the full year of 2025 (adjusted) is approximately $251 million, although it has narrowed compared to before, there is still a considerable distance from profitability.

However, whether the premium of the Sci Tech Innovation Board can be realized depends on whether MiniMax is regarded as a “scarce large model asset” or a “replaceable API commodity”.

If the commercialization of M3 is hindered and B-end growth falls short of expectations, the A-share market will not offer a premium, and the valuation of Hong Kong stocks will also be subject to reverse scrutiny.

At the same time, if the process of returning to A encounters obstacles or the valuation falls short of expectations, the psychological value of option incentives may be shaken, leading to the loss of core AI talents and dragging down model iteration, which is a more fatal blow than a drop in stock prices.

From the perspective of the industry landscape, MiniMax is still facing subtle changes in the competitive situation.

With stronger programming capabilities and B-end genes, Zhipu has taken a significant lead in market value; DeepSeek quickly seized API market share through aggressive price reductions; Giants such as Xiaomi and Alibaba continue to squeeze the survival space of independent startups through ecological subsidy models.

Under the pincer attack from all sides, the answer to whether MiniMax can hold its position is not optimistic.

4、 Conclusion
MiniMax is currently facing three challenges:

In the short term, whether the lifting of the ban on selling pressure in July can be digested by the market; Can the mid-term report deliver the answer of “high growth in B-end revenue+improvement in gross profit margin”; In the long run, it remains to be seen whether the gross profit margin under the price war and the financing channel of the Science and Technology Innovation Board can support the company to reach a turning point in profitability.

This company has a solid technological foundation, a global user base, and a leading open-source model M3, all of which are assets that other domestic AI startups find difficult to replicate.

It also faces a fundamental contradiction: relying on the C-end “dopamine business” to make money and support high investment, long-term AGI research and development. In today’s world where computing power costs continue to rise, this path is becoming more and more like walking a tightrope.

The price hike controversy is a warning bell, but it is not enough to negate the full value of MiniMax.

What is truly worth observing is: as the brilliance of technology is gradually overshadowed by the shadow of computing power costs, and as the patience of the capital market shifts towards real numbers, can MiniMax make the crucial leap from having technology to being able to make money?

Prior to this, it is wiser to remain cautious rather than blindly optimistic or pessimistic.