On August 26, MiniMax unveiled its first-half fiscal 2026 results. According to information from the Zhaofeng trading desk, both UBS and Jefferies promptly released research notes, maintaining their "Buy" ratings. The two investment banks believe that MiniMax is experiencing accelerating growth, narrowing losses, and ARR flexibility that has exceeded expectations.

Revenue is accelerating, with the open platform emerging as the primary growth engine. MiniMax's total revenue for the first half of fiscal 2026 reached $116.6 million, representing a year-on-year surge of 283%.

Breaking down the business segments: The open platform and AI enterprise services generated revenue of $73.9 million, up 703% year-on-year, with its share of total revenue jumping from 30% in the first half of fiscal 2025 to 63%. Growth drivers include an increase in paying users and enterprise clients, higher API call volumes, and the rapid adoption of token packages. Meanwhile, AI-native products brought in $42.6 million in revenue, up 101% year-on-year, benefiting from improved user engagement, stronger willingness to pay, and continued monetization of products such as Hailuo AI.

Jefferies further analyzed the geographical breakdown: revenue from mainland China grew 434% year-on-year to $46 million, accounting for 39% of total revenue; revenue from other regions climbed 224% year-on-year to $71 million, representing 61%. As of June 30, 2026, MiniMax had served more than 230 countries and regions worldwide.

Gross margin is under pressure, but improvement is expected in the second half. The gross margin for the first half of fiscal 2026 stood at 17.9%, up 5.8 percentage points year-on-year, yet still below the levels seen in the fourth quarter of fiscal 2025 and the full-year average.

Jefferies analysts Thomas Chong and Zoey Zong pointed out in their report that the gross margin was dragged down by three factors: a one-time compensation to customers in June; the M3 model requiring time to stabilize in its early stages, with cost optimization such as throughput enhancement still in progress; and the impact of promotional activities for token packages.

However, management remains optimistic about the second half. Jefferies quoted management's stance: "Gross margin will improve in the second half of the year, with further room for enhancement in 2027."

On the cost side, UBS noted that research and development expenses increased 139% year-on-year, but their ratio to revenue fell to 255%, down 154 percentage points year-on-year. Sales and distribution expenses declined 18% year-on-year, with their ratio dropping to 23%, down 85 percentage points, reflecting a significant contraction in promotional spending following the company's shift toward an organic growth strategy.

ARR has exceeded expectations, with July token consumption reaching 20 times the level at the start of the year. MiniMax's August ARR has already surpassed $800 million, well above Jefferies' previous forecast of $580 million.

A month-by-month breakdown of ARR drivers: In late June, M3 model token consumption continued to rise, with TPS (tokens per second) increasing from 20 to 100. In July, rapid growth in text model usage boosted ARR. In August, the release of the H3 model contributed incremental ARR.

July token consumption reached 20 times the January 2026 level. In terms of ARR structure, the To B (enterprise) segment contributed approximately 80%, compared to only 30% in the same period last year. The number of enterprise customers and developers has grown about 10-fold since the end of last year, now exceeding 2 million, with a particular concentration in agent-related task areas.

Upcoming model lineup: Jefferies' report detailed MiniMax's near-term model roadmap, including three upcoming models. The M3.1 aims to further strengthen pre-training and post-training to prepare for the next generation of large-scale models. The M3 Pro, with parameters approaching 3 trillion, features upgrades across architecture, inference efficiency, and latency. The H3.1 seeks to enhance model intelligence based on user feedback from H3.

Management stated that the core of model competition does not lie in the confrontation between internet giants and AI labs, but rather in model intelligence, unit cost efficiency, and commercialization capability. Regarding computing power, MiniMax indicated it has sufficient self-built infrastructure to support operations, while also collaborating with cloud service providers and beginning to use domestic chips, with expectations of further contributions in the fourth quarter.

UBS maintained its "Buy" rating with a 12-month target price of HK$500. At the current share price of HK$303, this implies a projected stock return of 65%, using a Price/ARR valuation method. Jefferies also maintained its "Buy" rating, raising its target price to HK$533 from the previous HK$506. With the current share price at HK$299.60, the target price implies an upside of approximately 78%. In the base case, the target price is HK$533, with an upside scenario of HK$639 and a downside scenario of HK$213.

UBS noted that due to limited consensus estimates available for the first half of fiscal 2026, investors on the earnings call will focus more on management's statements regarding the latest ARR trends, year-end 2026 targets, model iteration roadmap, and AI investment outlook.