China’s MiniMax 2025 revenue jumps 158% to $79m

MiniMax Group Inc, an AI foundation model company, reported FY2025 results for the year ended Dec 31, 2025, saying revenue rose 158.9% year-on-year to US$79.0 million and gross profit increased 437.2% to US$20.1 million.

The company said more than 70% of revenue came from international markets and it had cumulatively served over 236 million users and 214,000 enterprise customers and developers by year end.

Adjusted net loss, defined by the company as net loss adjusted for share-based payment expenses, fair value loss on financial liabilities, and listing expenses, was US$250.9 million in 2025, slightly wider than US$244.2 million in 2024.

The cash balance was US$1,050.3 million at year end, which the company said includes cash and cash equivalents, financial assets at amortised cost, financial assets at fair value through profit or loss, restricted cash, and time deposits.

MiniMax reported R&D spending rose to US$252.8 million while selling and distribution expenses fell to US$51.9 million and administrative costs rose to US$36.8 million.

The company said it released new model versions including the M2 series during 2025 and the M2.5 model in February 2026 and rolled out multimodal models such as video Hailuo 2.3 and speech 2.6.

MiniMax said its video models helped creators generate more than 600 million videos and its speech model generated over 200 million hours of speech.

🔗 Source: MiniMax

🧠 Food for thought

Implications, context, and why it matters.

The numbers hide a story of investor frenzy and disputed model-training allegations

MiniMax reported R&D spending of US$252.8 million, yet that number alone does not explain how its models were built.
US-based competitor Anthropic alleged that MiniMax created 13 million exchanges with its Claude model through unauthorized third-party application programming interface (API) resellers, raising questions about whether rival data distillation sped up MiniMax’s gains 1.
MiniMax, Moonshot, and DeepSeek now rank among the most-used options on OpenRouter (a service that lets developers route AI prompts to different models). Together they accounted for nearly two-thirds of token usage among the top five models on the platform, which tracks rising interest in Chinese open-source models 2.
Markets also reacted. MiniMax’s stock rose 488% after its January debut in Hong Kong as investors in China chased AI winners. That move ran against the “AI scare trade” seen in US markets 3.

A new playbook for AI competition emerges from China

MiniMax’s plan to move from a large-model company to a “platform company” fits a broader Chinese approach that targets performance, accessibility, and cost in competition with US AI leaders 4.
On third-party platforms, MiniMax lists pricing as low as $0.30 per 1 million input tokens, pushing advanced AI toward commodity pricing and undercutting some US rivals 5.
Its public listing also gives investors a direct way to buy into an AI-focused company, unlike the US where OpenAI and Anthropic remain private 3.
Lower prices plus public market access could draw global capital and developer talent toward China’s ecosystem, widening the contest beyond benchmark scores.