MiniMax H1 2026 Earnings: $116.6M Revenue, a $293.0M Adjusted Loss, and a New Balance Sheet

MiniMax generated $116.6 million of revenue in the first half of 2026—more than the $79.0 million it produced in all of FY2025—as Open Platform revenue surged 703%. Reported net loss improved 11% to $358.0 million, but the cleaner operating measure moved the other way: adjusted net loss more than doubled to $293.0 million as R&D spending reached $296.9 million. The business commercialized at extraordinary speed. It also spent almost every new dollar, and then some, pursuing the next model cycle.

The Headline Numbers​

MiniMax's interim results announcement reports the six months ended June 30, 2026 under IFRS, in US dollars:

The table contains two opposing stories. The operating engine strengthened: revenue grew faster than cost of sales, gross profit grew 465%, and selling expense fell despite a near-quadrupling of sales. But the investment engine accelerated too. R&D rose by $172.5 million, twice the $86.1 million increase in revenue. That is why adjusted loss worsened even as reported loss improved.

The distinction matters because the reported comparison is flattered by capital-structure accounting. H1 2025 included a $253.9 million fair-value loss on pre-IPO financial liabilities. H1 2026 included only $31.0 million after the convertible preferred shares became ordinary equity at the January listing. Remove that swing, share-based payments, and listing expenses, and MiniMax's underlying loss rose 111%. The top line is scaling. Profitability is not yet following it.

Revenue Deep Dive: The Platform Became the Company​

MiniMax sells the same foundation-model stack through two channels: AI-native applications such as Hailuo AI and Talkie, and an Open Platform used by developers and enterprises. H1 2026 changed which channel defines the company.

AI-native products still doubled. The filing attributes the gain to higher engagement, greater willingness to pay, and continuing monetization of Hailuo AI and other applications. That is a healthy consumer result on its own. It simply looks modest beside a platform business that became eight times larger in one year.

Open Platform added $64.7 million of revenue and supplied three quarters of MiniMax's total growth. Management tied the increase to more paying individuals and enterprise customers, higher API call volumes, and rapid adoption of the Token Plan. The company's results release adds the operating context: token consumption in July was 20 times its January level. The monetization mix has flipped from applications wrapping MiniMax models to customers paying directly for model inference.

Geography tells a subtler story. Revenue outside Chinese mainland rose from $21.858 million to $70.828 million, but its share fell from 71.8% to 60.8% because mainland revenue grew even faster, from $8.571 million to $45.745 million. MiniMax is not becoming less global; both regions expanded rapidly. It is becoming less dependent on international consumer apps as Chinese enterprises and developers adopt the platform.

This mix is strategically better than the FY2025 shape. A consumer application must repeatedly win attention and pay distribution costs. An inference platform embeds itself in customer workloads, and usage can compound with the customer's own growth. The falling selling expense supports that thesis. The risk is that platform revenue carries heavy compute costs and aggressive pricing expectations. A 703% growth rate proves demand, not durable unit economics.

The Margin Story​

MiniMax's margins show commercialization outrunning inference cost while training investment races ahead of both:

Gross margin expanded 5.8 percentage points year over year because revenue grew 283% while cost of sales grew 258%. Management attributes the improvement to infrastructure efficiency. That is important evidence that MiniMax can deliver more inference per dollar of serving cost. The 17.9% result remains below FY2025's 25.4%, however, so mix and rapid platform adoption have not produced a straight-line margin climb.

Selling efficiency is the cleanest positive signal. Selling and distribution expense declined by $5.9 million even as revenue added $86.1 million. MiniMax says lower promotional spending reflects an organic user-growth strategy. The ratio fell from 108% of revenue to 23% in a year. A model company that can distribute through developer adoption and product usage rather than paid promotion has a materially better path to scale.

R&D is the deliberate pressure point. Spending rose 139% to $296.9 million, driven mainly by cloud services for training as MiniMax increased model iteration and upgraded its language and multimodal systems. The ratio improved from 409% to 255% of revenue, but that is denominator progress, not a dollar reduction. H1 R&D alone exceeded all FY2025 revenue by 3.8 times. Gross profit cannot yet fund even one-tenth of it.

Administrative expense doubled to $30.2 million because of management headcount, share-based payments, and professional fees, yet fell to 25.9% of revenue from 48.8%. That is real operating leverage. It is simply overwhelmed by the size of the training program.

The One Big Question: Has Commercialization Caught the Training Bill?​

Not yet. MiniMax's growth rate is spectacular, but the incremental economics are still negative.

For every new revenue dollar, MiniMax generated about 20 cents of additional gross profit and spent about $2.00 more on R&D. The platform's commercial acceleration therefore did not finance the model-development acceleration. Adjusted loss grew by $154.3 million—about $1.79 for each incremental revenue dollar.

That does not make the spending irrational. Foundation-model competition rewards capability, inference efficiency, and release cadence before it rewards near-term earnings. MiniMax released M3 during the reporting period and H3 shortly afterward, while expanding coding, agentic, video, and multimodal workloads. A slower R&D plan could protect cash and lose the product edge that created 703% platform growth.

The falsifiable investment question is whether the relationship changes over the next two reports. Gross margin needs to move back toward or above FY2025's 25.4%, and R&D growth needs to remain below revenue growth in dollars as well as percentage terms. If annualized revenue approaches $250–300 million while R&D remains near the H1 run rate, the loss can begin to narrow. If each model cycle requires another step-up in cloud training expense, platform scale will keep funding only a fraction of research.

The balance sheet gives MiniMax time. Its defined cash balance was $1.323 billion at June 30, up from $1.050 billion at December 31, including cash equivalents, financial assets, restricted cash, and time deposits. After period end, the company also completed a share placement and a zero-coupon convertible-bond issue. Liquidity is not the immediate constraint. The question is whether that capital buys a self-funding inference business before competition resets model prices again.

Tracking a $1.7 Billion Balance Sheet in Plain Text​

Modeling MiniMax in Beancount makes the distinction between operating loss and financing transformation unavoidable: double-entry forces every amount to reconcile to a source and to the other side of the balance sheet.

Here is the H1 2026 income statement from the pushed ledger. Income accounts are negative credit balances, Expenses accounts are positive debits, and the negative equity adjustment offsets the filed net loss so the transaction sums to zero:

; FY2026 H1 Income Statement — six months ended June 30, 2026
; Check: −116.573 + (−6.288) + 95.760 + 296.870 + 57.203 + 31.025 + 0 + (−357.997) = 0 ✓
 
2026-06-30 * "MiniMax Group Inc." "FY2026 H1 Income Statement"
  Income:Revenue                         -116.573 MUSD
  Income:OtherNet                          -6.288 MUSD
  Expenses:CostOfRevenue                   95.760 MUSD
  Expenses:ResearchAndDevelopment         296.870 MUSD
  Expenses:SellingGeneralAdministrative    57.203 MUSD
  Expenses:FairValueLossOnFinancialLiabilities 31.025 MUSD
  Expenses:IncomeTax                        0 MUSD
  Equity:Adjustments                     -357.997 MUSD

The balance-sheet lines carry the larger narrative. Convertible redeemable preferred shares were a $3.598 billion liability at FY2025. They converted into ordinary shares when MiniMax listed on January 9, 2026, and the liability was zero by June. Filing-reported equity moved from a $2.648 billion deficit to positive $1.347 billion:

2026-06-29 pad Liabilities:Current:ConvertibleRedeemablePreferredShares Equity:Adjustments
2026-06-30 balance Liabilities:Current:ConvertibleRedeemablePreferredShares 0 MUSD
 
2026-06-29 pad Equity:CommonStockAndAPIC                 Equity:Adjustments
2026-06-30 balance Equity:CommonStockAndAPIC              -0.020 MUSD
 
2026-06-29 pad Equity:RetainedEarnings                   Equity:Adjustments
2026-06-30 balance Equity:RetainedEarnings            -1346.663 MUSD

This is why the reported net-loss improvement should not be mistaken for operating leverage. The income statement still contains $31.0 million of fair-value loss for the period, but the giant preferred-share liability is gone. Total liabilities fell to $368.0 million, total assets reached $1.715 billion, and the filing's gearing ratio collapsed from 343.3% to 21.5%. The company did not repay $3.6 billion in cash. The instrument changed classification from liability to equity.

The Multi-Year Arc​

MiniMax's five reported snapshots trace a company moving from a pre-revenue research lab to a listed inference platform:

Revenue has gone from zero to a H1 run rate above $230 million in four years. R&D has expanded almost as dramatically, from $10.6 million in FY2022 to $296.9 million in six months. That is the central bargain: the research program created a real platform business, but it has not yet become economically subordinate to that business.

The financing arc is just as important. MiniMax raised capital through preferred shares while private, and IFRS treated those shares as liabilities whose rising fair value produced enormous non-cash losses. Cash and short-term investments climbed from $70.5 million to $959.9 million by FY2025, while the preferred liability climbed to $3.6 billion. Listing converted the liability into equity, leaving H1 2026 with a conventional-looking capital structure and $1.223 billion of cash plus short-term investments.

H1 2026 is therefore the first clean starting point for judging the public company. Future results will no longer be dominated by the same preferred-share remeasurement. Investors will see the operating question directly: platform revenue and gross profit on one side, model-training expense on the other.

The Verdict: Bull vs. Bear​

Bull Case

• Revenue grew 283.1% to $116.6 million and exceeded the entire FY2025 result in six months.
• Open Platform revenue grew 703.1% to $73.9 million and became 63.4% of the mix, creating a usage-driven enterprise engine less dependent on consumer-app promotion.
• Gross profit grew 464.8%, faster than revenue, and gross margin expanded 5.8 percentage points as infrastructure efficiency improved.
• Selling and distribution expense fell 17.9% in absolute dollars while revenue nearly quadrupled, pushing the ratio from 107.9% to 23.1%.
• R&D growth of 138.8% was far below revenue growth in percentage terms, and R&D as a share of revenue fell from 408.6% to 254.7%.
• The preferred-share conversion repaired the balance sheet: $1.347 billion of positive equity, 21.5% gearing, and a $1.323 billion defined cash balance give the company time to fund model development.

Bear Case

• Adjusted net loss more than doubled to $293.0 million; the 11% improvement in IFRS loss came from a $222.9 million reduction in fair-value charges, not from operating profitability.
• Incremental R&D spending was twice incremental revenue and more than ten times incremental gross profit.
• H1 gross margin of 17.9% remains below FY2025's 25.4%, suggesting the fast-growing platform mix still carries substantial inference cost or pricing pressure.
• R&D consumed 255% of revenue, and H1 spending already exceeded FY2025's full-year R&D total by $44.1 million.
• Administrative expense doubled, while adjusted loss grew $1.79 for every incremental revenue dollar.
• Platform customers can scale usage quickly, but they also compare model quality and price continuously; the same market that produced 703% growth can compress inference economics just as fast.

Our Take: MiniMax proved product-market fit more convincingly in H1 2026 than in any previous period. A platform business growing 703%, a consumer business doubling, expanding gross margin, and falling selling expense is not a speculative revenue story anymore. But the filing does not yet prove operating leverage. The preferred-share conversion makes the balance sheet cleaner and the reported loss smaller; it does not make training cheaper. The next milestone is not another triple-digit revenue headline. It is a report where incremental gross profit funds a meaningful share of incremental R&D and adjusted loss starts to fall. Until then, MiniMax is a rapidly commercializing research program with ample capital—not yet a self-funding software business.