by ZHU Junxi
MiniMax, a China-based artificial intelligence model developer, reported a sharp rise in revenue in its first full-year results since listing, though accounting adjustments tied to higher valuations pushed net losses wider.
For 2025, the company posted revenue of $79.04 million, up 158.9% from a year earlier and above analysts' average forecast of $71.4 million.
Net loss widened to about $1.87 billion, compared with $465 million in 2024. MiniMax said the increase was mainly due to significant remeasurement losses on preferred shares as its valuation rose.
Excluding such items, adjusted net loss came in at $250 million, slightly higher than the $244 million recorded a year earlier.
More than 70% of revenue was generated overseas, according to the company.
Revenue from AI-native applications reached $53.1 million, accounting for 67.2% of total revenue and rising 143.4% year on year. Open-platform and enterprise AI services generated $26 million, up 197.8%, lifting their share of revenue to 32.8% from 28.6% in 2024.
Research and development expenses increased to $253 million in 2025 from $189 million a year earlier, reflecting continued upgrades of base models and multimodal capabilities as well as higher cloud costs associated with training. The company said R&D spending grew at a slower pace than revenue, indicating improved efficiency.
"In AI, success isn't about how much capital or computing power you deploy, but how fast your models improve," founder and chief executive YAN Junjie said on a conference call. "And that ultimately comes down to R&D efficiency."
MiniMax has rolled out several models since late October, including the M2, M2.1 and M2.5 series. The latest M2.5 model, launched in February, was positioned for coding and tool-use scenarios. While it trails Chinese peers such as Zhipu's GLM-5 and Alibaba's Qwen-3.5 in some comprehensive benchmarks, MiniMax said its lower activation parameters and pricing advantages have made it the most-used model on API platform OpenRouter.
Development of next-generation M3 and Hailuo 3 models is underway, Yan said, adding that the upgrades are designed to deepen multimodal integration.
The company expects advances in autonomous coding, office AI tools and multimodal content creation to drive token usage growth of one to two orders of magnitude in 2026, with the forthcoming models positioned to capture that demand.
MiniMax does not plan to compete in the general-purpose mobile assistant segment, instead focusing on enhancing model capabilities and building a full-modal AI ecosystem.
Since listing in Hong Kong on Jan. 9, MiniMax shares have surged more than 400%. The stock rose as much as 15% at the open on Tuesday and closed up 9.1% at HK$821.00, giving the company a market capitalization of about HK$257.5 billion.
