Moonshot AI pivots Kimi models from coding to science as benchmark gains flatten
Reported by The Decoder and MLQ.ai2 sources ↓
Moonshot AI says it is redirecting Kimi models toward biology, chemistry and pharmaceutical research as gains from coding benchmarks diminish. Its international director said science requires more capable AI than law, finance or code.
Key points
- DP Technology is integrating Kimi K3.1 Pro into its BioMaster research assistant
- In a test, the AI analyzed data from 3,696 mouse pancreatic cells and flagged 100 candidate genes in 90 minutes
- Leading models produced verifiable proofs on a theoretical computer science benchmark but solved none of its open conjectures
- A June survey found 24% of 50 US research leaders expect AI to noticeably improve drug development success rates
Sources
Details
The Decoder reports that Moonshot AI international director Giacomo Ficari described the shift at the Web Summit in Lisbon. Beijing-based DP Technology is integrating Kimi K3.1 Pro into BioMaster, a research assistant. In one test, the AI analyzed gene data from 3,696 mouse pancreatic cells and identified 100 candidate genes in 90 minutes; the article says the work was estimated to take a human four hours.
On TCSExam, a 100-problem theoretical computer science benchmark, leading models produced verifiable proofs but did not solve any open conjectures on their own; US competitors faced similar limits. It also notes that a June Citi survey of 50 US research leaders found 72% already use AI broadly, while 24% expect it to noticeably improve drug-development success rates.
Written from coverage as of