China's Moonshot AI and MiniMax Emerge as Top Frontier Lab Rivals
- tech360.tv
- 14 minutes ago
- 3 min read
Chinese artificial intelligence start-ups Moonshot AI and MiniMax are emerging as strong contenders to rival US frontier labs. They are positioned as China’s strongest challengers in 2025, even as DeepSeek has garnered significant attention.
Moonshot AI, founded by Yang Zhilin, has recently enhanced its presence in China’s AI ecosystem. This follows the launch of Kimi K2 Thinking, an upgraded reasoning model.
The system surpassed OpenAI’s GPT-5 and Anthropic’s Claude Sonnet 4.5, advanced closed-source AI models, on various benchmarks.
Deedy Das, of Menlo Ventures, described the Kimi K2 Thinking release as "a turning point in AI," noting a Chinese open-source model had achieved a top position. Nathan Lambert, of the Allen Institute for AI, also commended Kimi K2 Thinking for reducing the gap between open-source and leading closed-source systems.
MiniMax, led by founder Yan Junjie, has also gained global attention with its M2 model. The M2 model last month climbed to the top of a significant leaderboard for open models.
MiniMax M2 secured a record score for an open model on Artificial Analysis’s overall intelligence index. This placed it ahead of Google DeepMind’s Gemini 2.5 Pro and close behind the latest US models from OpenAI and Anthropic.
The emergence of Moonshot AI and MiniMax highlights China’s capacity to challenge the US in developing fundamental AI models. While tech giants like Alibaba Group Holding and ByteDance advance their own large language models and AI infrastructure, new Chinese start-ups are also progressing to compete globally.
An AI industry executive stated that China’s extensive talent pool is crucial for companies like Moonshot AI and MiniMax. These firms can outperform expectations despite having fewer high-end chips and less funding than many US counterparts.
Moonshot AI, for instance, trains models with notably fewer high-end graphics processing units (GPUs) than US rivals, yet maintains confidence in its development.
In a Reddit discussion in early November, an account believed to belong to Yang was questioned about the launch of the company’s next-generation foundational model, K3. The query referenced OpenAI’s substantial data centre ambitions.
Yang responded, "Before Sam’s trillion-dollar data centre is built."
Yan, the founder of MiniMax and a former computer vision specialist at SenseTime, focuses on multimodal models that process text, images, and video.
MiniMax is considered among the first in China to implement mixture-of-experts (MoE) architecture at scale. This approach was subsequently adopted and popularised by DeepSeek.
Moonshot AI and MiniMax’s latest open-source models have surpassed DeepSeek’s offerings in Artificial Analysis’ rankings, attracting investor interest.
MiniMax raised USD 300 million in a strategic round in July, achieving a valuation over USD 4 billion. The company has since filed for a Hong Kong listing, aiming to raise up to HKD 5 billion.
Moonshot AI also secured new funding, raising approximately USD 600 million in a funding round last month. This round was led by Beijing-based venture capital firm IDG Capital and Tencent Holdings, according to ITJuzi.com.
Analysts caution that these start-ups will face strong competition from well-funded Big Tech players. Wang Sheng, an investor at InnoAngel Fund, noted the difficulty for AI start-ups to flourish when directly competing with large technology companies.
"AI start-ups have to either jump the gun or position themselves differently," Wang said. "Otherwise it will be difficult for them to win the same game."
Moonshot AI and MiniMax are emerging as strong rivals to US frontier AI labs.
Moonshot AI’s Kimi K2 Thinking and MiniMax’s M2 models have outperformed established AI systems on benchmarks.
Both start-ups have recently secured substantial funding rounds, attracting significant investor interest.
Source: SCMP