Moonshot AI’s Kimi Model Is A Research Engine Focusing On Scientific Discoveries


China’s Moonshot AI is turning its focus to further scientific discovery, joining its US counterparts in a race to employ AI as a research engine. Moonshot AI has developed the Kimi models and is exploring the possibilities of developing advanced AI reasoning capabilities, particularly in the fields of biology, chemistry and pharmaceutical research.
According to Giacomo Ficari, Moonshot’s Director of International, science is a fascinating field to push the limits of AI. He explained that finance and law are at the same intelligence level as coding in terms of AI capabilities, but it is science which has the potential to push the level of intelligence even higher. Ficari said that despite growing demand for AI models, technological progress was plateauing, prompting Moonshot to explore applications in scientific and pharmaceutical research. “You see OpenAI focusing on mathematics, and we are doing the same in biology and chemistry”, he said.
To actualise its plans, the start-up recently announced a partnership with DP Technology – a Beijing-based AI-for-science start-up. The two companies will integrate the Kimi K3.1 Pro model to further research in biomedical applications. DP Technology said that Kimi K3.1 Pro scored the highest rank in an internal evaluation testing AI models for biological research tasks.
One experiment included BioMaster, DP Technology’s research agent, and Kimi analysing gene-expression data from 3,696 mouse pancreatic cells. Of these, 100 genes were identified as needing further investigation. The entire process took only 90 minutes, compared to four hours of human work. However, despite their time-saving nature, these advanced AI models still cannot make autonomous scientific breakthroughs.
Earlier this month, Moonshot creator Yang Zhilin and Turing Award-winning computer scientist Andrew Yao co-authored a study that highlighted this distinction and looked at whether cutting-edge AI models may promote new mathematics research. The study comprised a benchmark set of 100 tasks that included attempts to address long-standing unsolved problems in theoretical computer science, algorithm development, and mathematical theorem proving.
The researchers discovered that whereas frontier models could help with difficult research tasks and generate verifiable results, none of them could independently solve the benchmark's open mathematical conjectures, underscoring both the potential and present limitations of AI in autonomous research.
This is a new turn in the US-China AI race, with AI models being used to solve longstanding scientific or mathematical problems and conundrums and make discoveries. Moonshot’s announcement comes only a week after OpenAI confirmed in a blog post that its models had made three discoveries across mathematics, computer science and physics, generating proofs for long-standing problems. However, the results garnered much backlash, particularly from mathematicians. Similarly, in September, Anthropic’s researchers had used its Claude model to discover a new enzyme.
However, researchers are exercising caution while exploring pharmaceutical R&D opportunities, as they remain uncertain of the near-term impact of AI. A survey conducted by Citi in June on 50 US-based pharmaceutical R&D executives showed that only 24% believed AI could significantly improve the probability of drug success, despite a whopping 72% confirming that their firms were in the process of integrating AI into their research operations.
It remains undisputed that AI is here to stay and will be integrated into all industries. However, before becoming fully engulfed by the AI wave, entrepreneurs must confirm the utility, cost and impact of such technology, rather than blindly following market trends. It is a positive sign that the pharmaceutical industry is being cautious with AI usage, but perhaps newer models like Kimi can help further research and success in these industries, thus making AI an indispensable asset.





