AsianFin -- China's fierce battle for large AI model supremacy has entered a new phase as upstarts MiniMax and Moonshot AI go head-to-head with DeepSeek, the new heavyweight in China's open-source AI race.

With both companies launching new models and products in rapid succession, the rivalry reflects intensifying competition among the so-called "Six Tigers of Large Models" — MiniMax, Moonshot, Zhipu AI, Baichuan, 01.AI, and StepStar, all of which are now eyeing public listings amid funding pressures and deep-pocketed rivals.

MiniMax made headlines last week with the release of five new products in five days, including its flagship inference model MiniMax-M1. Priced at just $530,000 for reinforcement learning training, M1 matches DeepSeek in performance and offers the longest context window in the market.

Other rollouts included its video model Hailuo 02, general-purpose MiniMax Agent, Hailuo Video Agent, and the updated MiniMax Audio. Meanwhile, Moonshot AI debuted Kimi-Dev-72B, an open-source code model with 72 billion parameters that beat DeepSeek on the SWE-bench Verified benchmark with a 60.4% accuracy rate. The company also launched its first intelligent agent, Kimi-Researcher, for deep research tasks.

These back-to-back launches mark the first direct technological confrontation between Moonshot CEO Yang Zhilin and MiniMax's Yan Junjie, both targeting DeepSeek's dominance. Analysts see the moves as an attempt by the smaller players to remain visible in an increasingly crowded AI field dominated by ByteDance, Alibaba, and DeepSeek.

MiniMax is reportedly planning an IPO in Hong Kong with a \$3 billion valuation, following in the footsteps of Zhipu AI, which began IPO counseling with CICC in April. According to Alphabet List, five of the Six Tigers are preparing to go public as competition for funding intensifies. However, despite a burst of innovation, the Six Tigers are grappling with capital shortfalls, talent attrition, and uncertain paths to commercialization.

Recent data from IDC projects China's generative AI software market will reach just RMB 4.89 billion by 2028, underlining concerns about monetization and market scale. As large AI models transition from foundational capabilities to inference services, companies are racing to find viable business models amid heavy R\&D costs and limited consumer application uptake.

"Before DeepSeek, people assumed Chinese models lagged the U.S. by years. Now we know the gap is just a few months," said one analyst, referencing remarks by U.S. policy advisor David Sacks. Still, DeepSeek's meteoric rise has shaken investor confidence in its smaller rivals. Several Six Tigers have seen senior executives depart for larger firms in recent months, including Zhipu, MiniMax, and 01.AI.

Zhipu, which saw four executives exit in 2025 alone, is nonetheless pushing ahead with its listing. Its commercial platform has grown ARR over 100% since the start of the year, with API revenue up 30-fold and over 25 million users on its Qingyan assistant.