10x Genomics Adds Claude-Powered Sentira Analysis to Chromium

10x Genomics is putting AI-assisted analysis directly into its single-cell workflow, taking aim at a persistent bottleneck: the custom coding that often separates Chromium data from answers researchers can use. Its new cloud application, Sentira Single Cell, uses Anthropic’s Claude models to help users explore data through a chat interface. The promise is faster analysis and broader access; the harder question is how researchers will verify results that can look convincing even when they’re wrong.

From custom scripts to conversational analysis

Single-cell experiments produce data that researchers examine through methods tailored to their scientific questions. That variety has made it difficult to build software that covers every use case, says Michael Schnall-Levin, PhD, CTO of 10x Genomics. The result, in his description, is a “last mile problem”: researchers frequently need to write or adapt analysis code themselves, a task that can require bioinformatics expertise.

Sentira is designed to let Chromium users ask questions about their data without writing that code directly. Behind the chat interface, the system generates scripts and runs analyses on 10x Cloud. Researchers can ask follow-up questions as they inspect results, while the service handles the computing infrastructure and can run multiple analyses in parallel.

Schnall-Levin estimates that work which once took weeks could take hours. He sees the potential benefit extending beyond faster individual projects: easier access to analysis could help more researchers engage directly with single-cell data, and automated exploration may surface patterns that users otherwise overlook. Those are plausible gains, but they depend on how well the application handles questions that vary from one experiment to another.

Sentira is 10x Genomics’ umbrella for software capabilities. The company launched Sentira Spatial a few weeks before the single-cell announcement; the new application is specifically for single-cell analysis. The platform can work with different large language models, though 10x says Anthropic’s Claude models are performing well for this use. The company is not training its own model.

A project shaped by the virtual lab idea

Work on Sentira began at the start of 2026, in collaboration with James Zou, PhD, an associate professor in Stanford University’s Department of Biomedical Data Science. Zou’s work on the “Virtual Lab” frames scientific research as a process that algorithms can help conduct. The concept appeared in a 2025 Nature paper, “The Virtual Lab of AI agents designs new SARS-CoV-2 nanobodies.” Sentira applies agent-based AI to a more routine but widespread research task: interpreting single-cell data.

10x says it will use researchers’ data to assess the application’s performance and improve Sentira, but that data will not feed back into the underlying Claude models. The distinction matters to labs weighing a hosted analysis tool: the company describes the data as supporting product improvements, not model training.

There’s also a competitive test ahead. Henry Long, PhD, scientific director of the Center for Functional Cancer Epigenetics at Dana-Farber Cancer Institute, has used both Sentira and Claude Science. He says a challenge for 10x will be keeping Sentira’s capabilities current alongside biomedical analysis tools from general-purpose AI companies, including Anthropic’s Claude Science and OpenAI’s Rosalind. Long says his lab has increasingly relied on Claude Science lately.

Credits put a price on analysis sessions

Sentira is sold through 10x Cloud credit packs. The company estimates that a typical agentic session will use about 12 credits, while a complete analysis through to experimental conclusions may take around 60. At launch, users receive 100 promotional credits for their first Chromium dataset uploaded to Sentira Single Cell—enough, according to 10x, to try the platform and reach experimental conclusions.

Pack prices range from $120 for 100 credits to $25,000 for 25,000. The per-credit price falls with larger purchases, from $1.20 to $1.00. On those estimates, a 60-credit analysis would cost $72 at the smallest pack’s rate, or $60 at the largest pack’s rate. Actual credit use will depend on how researchers work through their data.

Automation raises the bar for review

The central risk is not necessarily an obviously broken analysis. Luciano Martelotto, PhD, director of global market development for single cell at Waters Biosciences, has warned about AI analysis in the field more broadly: “nothing in an AI-generated analysis looks broken. The code runs, the plots are pretty, the story makes sense. The mistakes that matter are the silent ones.” In his view, a model can follow instructions and still lack the researcher’s instinct for questioning a result that looks too good.

Schnall-Levin says 10x considers Sentira useful for initial analysis and exploration, while stressing that it isn’t perfect. Researchers remain responsible for reviewing work before relying on it or including it in a publication. His concern is that increasingly capable tools could encourage people to accept outputs without thinking through their assumptions and limits.

Holger Heyn, PhD, an ICREA professor at the Centro Nacional de Análisis Genómico, sees AI helping researchers analyze millions of cells, automate cell annotation and identify patterns at scale. But he cautions that current systems haven’t captured the full complexity of human biology, in part because they lack sufficient human data in the loop. Experimental models remain valuable, he says, while representing only part of the cellular diversity and variation found in human health and disease.

There’s a longer-term tradeoff, too. If analysis becomes easier to outsource, some researchers may never build the skills needed to do it manually. Schnall-Levin compares that shift to modern navigation: few people need to find their way by the stars, and the skill has faded. Sentira could make single-cell analysis accessible to more people, but its practical value will rest on users knowing when to trust an output, when to question it and when to bring deeper analytical expertise to bear.