Artificial Intelligence and science have always been a combination with enormous potential, but one that rarely found a straightforward path to working in practice.
Anthropic just changed that game in a pretty significant way with the official launch of Claude Science, a work platform built specifically for researchers who need much more than a sophisticated text assistant. 🔬
The pitch is straightforward: bring together in a single environment all the tools and packages a scientist typically uses scattered across dozens of different platforms, produce auditable results, and offer flexible access to computing resources.
And the practical outcome is impressive.
Analyses that used to take years of intensive work are now being completed in a fraction of the original time, as you will see in the real-world cases we brought here.
But before diving into the details, it is worth understanding the context: why did this platform arrive now, what does it do differently, and why are researchers from institutions like the Allen Institute, UCSF, and companies like Manifold Bio already using it to push cutting-edge research forward? 👇
The Real Problem Claude Science Solves
Anyone who works in scientific research knows the daily routine all too well: you need to jump between dozens of databases, each with its own schema, deal with file formats that require custom-built data pipelines and specialized viewers, and still switch between an enormous list of tools like PubMed, Jupyter, R, cluster terminals, and so on. This fragmented workflow is not just inconvenient — it is one of the biggest real obstacles to the advancement of modern science. Time that could be spent thinking, forming hypotheses, and interpreting results ends up being consumed by friction between tools that simply do not talk to each other. And this repeats every single day, in labs and research centers around the world.
Receive the best innovation content in your email.
All the news, tips, trends, and resources you're looking for, delivered to your inbox.
By subscribing to the newsletter, you agree to receive communications from Método Viral. We are committed to always protecting and respecting your privacy.
Anthropic identified this pattern and built Claude Science specifically to eliminate that barrier. The platform brings all those fragmented tools together in a single research environment where scientists can carry out every step of their work. It helps analyze literature, run multi-step research workflows, produces detailed artifacts, and lets you iteratively refine figures and manuscripts until they are ready for publication. Just like in a Jupyter Notebook, you can access Claude Science where you already work — locally on macOS or Linux, or on a remote machine via SSH or through an HPC login node.
What makes this approach different from simply using a generic language model to help with scientific tasks is the level of specialization baked into the platform. Users interact with a generalist coordinator agent that has access to more than 60 skills and connectors carefully selected and pre-configured for genomics, single-cell, proteomics, structural biology, cheminformatics, and much more. These agents can create other agents and even communicate with specialist agents built by the users themselves. And there is more: a reviewer agent checks citations and calculations, flagging and correcting errors automatically. This is not a chatbot that answers biology questions — it is a tool that actively participates in the scientific investigation process. 🧬
Claude Science is launching today in beta for users on the Claude Pro, Max, Team, and Enterprise plans, and the platform will continue to be refined as Anthropic collects feedback from people using it in practice.
Rich, Fully Reproducible Scientific Artifacts
One of the features getting the most attention from early Claude Science users is the ability to generate scientific artifacts with an extremely high level of detail and, most importantly, with full reproducibility. Scientific research is inherently visual, so the platform generates figures and manuscripts side by side with the code that created them. It natively renders rich scientific artifacts, including 3D protein structures, genome browser tracks, chemical structures, and much more.
When the system generates a figure, it includes the exact code and environment that produced it, a plain-language description of how it was created, and the full message history. This lets you understand the input data, making the work easier to validate and reproduce, even months later. You can ask Claude Science to make edits to figures using natural language — like removing gridlines or switching an axis to log scale — and the agent edits the underlying code to fulfill the request. This level of transparency is something the scientific community values deeply, because it allows the researcher to maintain intellectual control over the process rather than simply trusting the output of an automated system blindly. 📊
Compute Management and On-Demand Scaling
Large analyses — like folding a protein or running a genomics pipeline over a massive dataset — usually require the researcher to shift focus to setting up the compute job, waiting while it gets submitted to a cluster, checking whether it succeeded or failed, and then retrieving the results. Claude Science handles that entire process for you. It drafts a plan, asks permission before accessing new resources, and lets you review or revoke any decision before writing and submitting the job to the compute resources your lab already uses — whether that is your own HPC cluster via SSH, or your Modal account for on-demand compute, scaling the analysis from a single GPU to hundreds as needed.
Because the agents work within an active session that maintains context in memory, even massive datasets only need to be loaded once. The platform runs on your own lab infrastructure — your laptop, Linux machine, or HPC login node — so large or sensitive data never needs to leave the systems where it already lives, and only the context required for each step of the analysis is sent to Claude. As the pipeline runs, a reviewer agent inspects the results, flagging incorrect citations, numbers without traceability, and figures that do not match the code that generated them, correcting itself along the way. You can also branch the session at any point to compare two approaches without losing the original thread. 🚀
Domain-Ready From Day One
Scientific knowledge is spread across hundreds of specialized sources. In biology, for example, relevant data can be distributed across resources like UniProt, PDB, Ensembl, Reactome, ClinVar, ChEMBL, and GEO — each with its own schema and query language — plus journals, preprint servers, and domain-specific open models. When you ask Claude Science a question in natural language, specialist agents query and synthesize information from all of these sources so you do not have to navigate them individually. The platform uses skills from NVIDIA’s BioNeMo Agent Toolkit to natively connect to BioNeMo’s life sciences models and libraries, including Evo 2, Boltz-2, and OpenFold3.
Scientists already have models, datasets, and pipelines they trust. Claude Science can also connect to those resources, saving any pipeline as a reusable skill or accessing your lab’s preferred tool through a connector, with future sessions inheriting everything automatically. This customization capability lets you access Claude, your proprietary data, and the validated tools you already trust within a single conversation. 🤝
What Scientists Are Already Doing With Claude Science
Over the past few months, researchers have been working with Claude Science in beta on tasks like single-cell RNA sequencing analysis, CRISPR screen design, protein structure prediction, cheminformatics, and much more. The real-world cases show just how big the impact is.
Manifold Bio, which designs tissue-targeted therapeutics — drugs that home in on a specific organ or cell type to act exactly where needed while sparing the rest of the body — used Claude Science to nominate targets for their most recent experiments. For each tissue and target, the platform evaluated surface expression, cellular trafficking, and safety, ranking candidates according to criteria the company learned from its own proprietary internal data. According to Manifold, what set Claude Science apart from a generic coding assistant was the ability to do this end to end, pulling together the right data and applying the right judgment with context from prior programs already built in.
Jérôme Lecoq, a neuroscientist at the Allen Institute, used the platform to build a multi-agent computational review template composed of around 20 custom skills geared toward writing long-form reviews. Sub-agents read thousands of papers, extracting the central claim and the key quantitative finding from each one, storing everything in an evidence database. The pipeline then constructs a narrative arc, writing the review section by section and delegating each part to a specialized sub-agent. A key component of the workflow is the use of actor-critic pairs: one agent creates the content while a separate reviewer agent evaluates accuracy and citation fidelity. Before Claude Science, writing a review like this could take up to two years for Lecoq’s team. Today he already has around 10 reviews, many exceeding 100 pages, with citations checked by reviewer agents.
And Stephen Francis, an associate professor and epidemiologist at UCSF‘s Brain Tumor Center, used Claude Science to support studies on the molecular epidemiology of glioma — a type of primary tumor that starts in the glial cells of the brain. His lab investigates how thousands of small-effect germline variants combine to shape individual susceptibility. Although this work existed before the platform, Francis said the app dramatically accelerated the analysis, enabling complete germline investigations in roughly one-tenth the time it used to take. His group independently validated the results, confirming that Claude Science can produce analyses that are both fast and robust. 🔬
How to Get Started With Claude Science
Claude Science is available in beta on macOS and Linux for the Pro, Max, Team, and Enterprise plans. Anthropic is releasing the tool early specifically so scientists can start using it on real problems and help refine the platform. Users on Team and Enterprise plans will need their administrator to enable Claude Science, and there is a Team plan with discounted seats for active scientific labs at academic institutions and nonprofit research organizations.
The company will also support up to 50 projects through its AI for Science program, offering up to 30 thousand dollars in credits, while Modal is providing up to 2 thousand dollars in compute for selected projects, with an initial focus on biology and biomedical research.
Why This Matters for the Future of Science
The launch of Claude Science comes at a time when artificial intelligence is moving from a distant promise to a concrete part of the global scientific infrastructure. What Anthropic is proposing with this platform is a step beyond: not just using AI as a specialized tool for specific tasks, but creating an integrated environment where artificial intelligence works as an active partner throughout the entire scientific research cycle — from literature review to the interpretation of final results.
This move has important implications for the speed at which new knowledge can be generated and validated. When the time needed to run complex analyses drops from years to weeks, or from weeks to days, the pace of science changes fundamentally. Researchers can test more hypotheses, course-correct more quickly, and devote more creative energy to the questions that truly require human intuition and original reasoning. AI, in this scenario, does not replace the scientist — it amplifies the scientist’s ability to do real science. And that is a distinction worth repeating, because it completely changes how we should think about the role of these technologies in academic and research settings.
Claude Science is still in its early adoption phases, but the results already being reported by leading institutions suggest that Anthropic has built something with real potential for lasting impact. As more researchers and institutions begin integrating the platform into their workflows, it will be possible to better understand how AI-mediated scientific data analysis will reshape not just individual productivity, but the entire logic of how science is organized, shared, and advanced collectively. And that is a story that is just getting started. 🔭
