Starts Nov 23Live Cohort 3 of Vibe Coding & AI Agents: 3 weeks, 9 live hands-on sessionsCohort 3 (Nov 23) →

BioMaster Agent on Kimi Picks 100 Genes in 90 Minutes Instead of 4 Hours

Moonshot AI, the maker of Kimi, believes the market for AI coding tools is nearing saturation and is moving into science. In 90 minutes, the BioMaster agent on Kimi K3.1 Pro selected 100 genes for further study from data on 3696 mouse cells. A human would need about four hours. But 72% of pharma companies already use AI, and only 24% expect it to raise drug success rates. Faster routine work is proven; "AI will invent the drug" is not, at least not yet.

AI-processed from TG: @ai_machinelearning_big_data (Machinelearning); edited by Hamidun News
BioMaster Agent on Kimi Picks 100 Genes in 90 Minutes Instead of 4 Hours
Source: TG: @ai_machinelearning_big_data (Machinelearning). Collage: Hamidun News.
Listen to article
Article Fact Sheet & Key Figures
ParameterDetails
Primary sourceTG: @ai_machinelearning_big_data (Machinelearning) ↗
Editor-in-Chief
BioMaster100 candidate genes from 3696 cells in ~90 minutes
Citi survey72% of 50 R&D executives already use AI
Key metric #3Only 24% expect a notable increase in the odds of success for new drugs

China's Moonshot AI, the maker of the Kimi models, says that programming, the main source of demand for frontier AI models, is nearing saturation. The company is looking for a new field in biology, chemistry and pharmaceuticals. It already has a first proof point: the BioMaster agent, built on Kimi K3.

1 Pro, took data on which genes are active in 3696 mouse pancreatic cells and, in about 90 minutes, selected 100 genes that a biologist should check next. By the company's estimate, a human would spend about four hours on the same selection. Picture a huge table where the rows are cells, the columns are genes, and each cell holds an activity level: the agent replaces the analyst who would otherwise hunt for patterns in it by hand.

For developers and founders, this is a signal: the race for the best model is shifting from the code editor to the lab, but pharma companies remain cautious about it.

Why Moonshot Is Moving Beyond Code

The company's head of international development spoke about this at the Web Summit conference in Lisbon. According to him, programming, which has driven most of the demand for frontier models, is approaching technological saturation: for most coding tasks, current model quality is already enough.

From here, by his logic, the bar goes up. Tasks in finance and law require roughly the same level of AI as work with code, and science raises it higher still. In science, it is not enough to write plausible text: the result has to survive experimental verification.

What Has Already Been Done: BioMaster on Kimi K3.1 Pro

Beijing-based DP Technology announced a partnership with Moonshot: Kimi K3.1 Pro will power its BioMaster agent, which is designed for biological tasks. It is an example of the "general-purpose model plus specialized agent" pattern: the model does the reasoning, and the agent connects it to the data and tools of a specific domain.

In a demonstration, BioMaster and Kimi analyzed gene expression, that is, which genes in a cell are "switched on" and how actively. From data on 3696 cells, it picked 100 candidate genes for further study in about 90 minutes, versus an estimated four hours of human work.

It is important to understand what exactly got faster. This is not drug discovery but preliminary screening: taking thousands of features and narrowing them to a hundred that a biologist will then test in the lab. This kind of routine eats up a lot of researchers' time. That said, the four hours is an estimate, not a measurement against an independent group, and the material does not report anything about the quality of the selected genes.

Pharma Is Skeptical

Developers' enthusiasm does not match the mood of the industry. A June Citi survey of 50 heads of R&D units at US pharmaceutical companies showed a gap between use and expectations. In 72% of the companies, AI is already used in research processes or fully deployed in them. But only 24% of respondents expect the technology to significantly raise the odds of success for new drugs.

In other words, AI can speed up data analysis, but no one is yet willing to guarantee that more drugs will reach patients. Pharma's main bottleneck is clinical trials, and a language model does not remove it on its own.

What to Do in Practice

For teams building agents, scientific domains open up a separate niche: the value comes from pairing a strong model with tools for a specific field. Product teams should sell what can be measured, such as saving time on exploratory analysis, 90 minutes versus four hours, rather than the promise that "AI will invent the drug," which the industry is not buying yet.

Moonshot's strategy is clear and other labs may repeat it: when the code market saturates, go where the cost of error is higher, and therefore so is the willingness to pay for quality. Whether Kimi holds its place in science will become clear when independent verification of BioMaster's results appears, not just a demonstration.

Story

Key Questions Answered

Why Moonshot Is Moving Beyond Code?

The company's head of international development spoke about this at the Web Summit conference in Lisbon. According to him, programming, which has driven most of the demand for frontier models, is approaching technological saturation: for most coding tasks, current model quality is already enough. From here, by his logic, the bar goes up. Tasks in finance and law require roughly the same level of AI as work with code, and science raises it higher still.

What Has Already Been Done: BioMaster on Kimi K3.1 Pro?

Beijing-based DP Technology announced a partnership with Moonshot: Kimi K3.1 Pro will power its BioMaster agent, which is designed for biological tasks. It is an example of the "general-purpose model plus specialized agent" pattern: the model does the reasoning, and the agent connects it to the data and tools of a specific domain. In a demonstration, BioMaster and Kimi analyzed gene expression, that is, which genes in a cell are "switched on" and how actively.

What are the key details on Pharma Is Skeptical?

Developers' enthusiasm does not match the mood of the industry. A June Citi survey of 50 heads of R&D units at US pharmaceutical companies showed a gap between use and expectations. In 72% of the companies, AI is already used in research processes or fully deployed in them. But only 24% of respondents expect the technology to significantly raise the odds of success for new drugs. In other words, AI can speed up data analysis, but no one is yet willing to guarantee that more drugs will reach patients.

ZK
Zhemal Khamidun — CPO at Alpina Digital, product manager of AlpinaGPT. Building AI products since 2019. Hamidun News is a pet project that grew into a production aggregator with 6,000+ readers.
✦ Unified AI Workspace · Nomad

Test Frontier LLMs & Autonomous Agents in One Workspace

Run Claude Opus 5.5, GPT-6, Gemini 3.1, and DeepSeek side-by-side in Nomad AI (nomadnet.ai) — unified model routing, deep research, and multimodal workflows under one subscription.

Try Nomad AI →Launch Nomad Agent
What do you think?
Loading comments…