Moonshot AI is changing course for its Kimi models: instead of chasing coding benchmarks, the Beijing startup wants to turn Kimi into an engine for scientific discovery, focused on biology, chemistry and pharmaceutical research.
The company's international director, Giacomo Ficari, made the case at Web Summit in Lisbon last week. In his view, science demands smarter models than law, finance or code — and gains in coding are already flattening, with coding benchmarks approaching saturation.
Partnership with DP Technology and the mouse-cell test
To put the plan into practice, Moonshot announced a partnership with DP Technology, a Beijing-based AI-for-science startup. The Kimi K3.1 Pro model will be integrated into BioMaster, the company's biomedical research assistant. According to DP Technology, Kimi K3.1 Pro scored highest in an internal evaluation of AI models for biological research tasks.
In a test cited by the company, the system analysed gene-expression data from 3,696 mouse pancreatic cells and flagged 100 candidate genes for investigation in 90 minutes — a task the company estimates would take about four hours of human work.
Independent breakthroughs are still far away
The enthusiasm comes with caveats. Frontier models still do not make independent scientific discoveries: on TCSExam, a benchmark of 100 theoretical computer science problems, the leading models produced verifiable proofs but did not crack any open conjecture on their own. The study was co-authored by Moonshot founder Yang Zhilin and Turing Award-winning scientist Andrew Yao.
The context is a race: in September, Anthropic's Claude found an enzyme system, while mathematicians criticised OpenAI's claimed breakthroughs in mathematics and physics. And the pharmaceutical industry remains sceptical: a June Citi survey of 50 US research leaders found that 72% already use AI broadly, but only 24% expect it to noticeably improve drug-development success rates.
Featured image: Kimi logo (2025), by Dontknowsalf, licensed under CC BY-SA 4.0, via Wikimedia Commons.
Sources
- The Decoder — Moonshot AI pivots Kimi models from coding to science as benchmark gains flatten (Nov 16, 2026)
- Oossa — Moonshot AI shifts Kimi toward biology and drug research (Nov 16, 2026)
- The Global Economics — Moonshot AI’s Kimi Model Is A Research Engine Focusing On Scientific Discoveries (Nov 17, 2026)
- South China Morning Post — “China’s Moonshot AI puts Kimi models into race for scientific discoveries” (original report)
Transparency: this article was produced with the help of artificial intelligence, cross-checking public posts and sources on the internet. Check the original sources to confirm the details.
By Muse