(Jan 27): Alibaba Group Holding Ltd-backed Moonshot AI released an upgrade of its flagship model, heating up a domestic arms race ahead of an expected rollout by Chinese sensation DeepSeek.
The latest iteration of Moonshot’s Kimi can process text, images and videos simultaneously from a single prompt, the company said in a statement on Tuesday (Jan 27), aligning with a trend toward so-called omni models pioneered by industry leaders like OpenAI and Alphabet Inc’s Google.
The model known as K2.5 joins a flurry of upgrades rolled out over the past month, reflecting a rush by China’s biggest AI players to get in ahead of the next unveiling by DeepSeek, which has teased a major release in recent weeks. Its research lab has published papers by prominent staff including co-founder Liang Wenfeng as well as code on GitHub.
Seeking to capitalise on reignited investor appetite for the country’s top-tier model makers, Moonshot last month raised US$500 million from investors including Alibaba and IDG Capital at a post-money valuation of US$4.3 billion, according to people familiar with the matter. To satisfy pent-up investor demand, the company has kicked off new rounds of financing seeking a valuation of as much as US$5 billion, one of the people said.
That comes after recent initial public offerings by Chinese AI rivals Zhipu and MiniMax Group Inc collectively raised more than US$1 billion in Hong Kong.
The three firms are among a handful of front-runners in a hotly contested battle among Chinese large language model makers, which at one point was dubbed the “War of One Hundred Models”. Following the breakout success of DeepSeek’s R1 model at the start of 2025, many smaller players have struggled to catch up with the requisite tech upgrades and funding.
See also: Musk’s X probed by EU over Grok’s spread of sexual deepfakes
Zhipu in January rolled out an image generation model GLM-Image that it says is the country’s first to be fully trained on domestic chips. Alibaba on Monday launched a reasoning version of its top proprietary model Qwen3-Max. The next day, its fintech affiliate Ant Group Co announced a spatial perception model for robots developed by Robbyant, a subsidiary.
K2.5 outperforms its open-source peers in several benchmark tests while narrowing the gap in coding with top-tier proprietary models, Moonshot said, adding that it is also rolling out an automated coding tool designed to rival Anthropic PBC’s popular Claude Code.
Moonshot was founded by former Tsinghua University professor Yang Zhilin, who had earlier stints working on AI projects at Meta Platforms Inc and Google. The company sells tiered subscription plans for its chatbot and offers its underlying technology to enterprise clients but trails peers like Zhipu and MiniMax in commercialisation.
Uploaded by Arion Yeow
