Anthropic released Claude Science on 30 June 2026, an app that pulls the scattered tools of research into a single workbench. For Australian universities, medical research institutes, biotech and agritech teams, the interesting part is not the demo. It is whether a structured research environment can replace the patchwork of scripts, subscriptions and cluster logins that most labs currently run.
What Claude Science actually is
It is a beta app for macOS and Linux, available to Claude Pro, Max, Team and Enterprise users (Team and Enterprise admins have to switch it on). You run it where you already work: on a laptop, a Linux box, or a remote machine over SSH or an HPC login node. At the centre is a coordinating agent with access to more than 60 curated skills and connectors, pre-configured for genomics, single-cell, proteomics, structural biology and cheminformatics.
Instead of navigating each source by hand, a researcher asks a question in plain language and specialist agents query across databases like UniProt, PDB, Ensembl, ClinVar and ChEMBL. Claude Science also connects to NVIDIA's BioNeMo Agent Toolkit, reaching life-sciences models such as Evo 2, Boltz-2 and OpenFold3. A reviewer agent sits alongside the work, checking citations and calculations, flagging untraceable numbers and figures that do not match their code, and correcting as it goes.
The feature that matters for R&D funding: auditable artifacts
The detail worth drawing out for Australian teams is reproducibility. When Claude Science produces a figure, it ships the exact code and environment that made it, a plain-language description of how it was built, and the full message history. You can annotate a figure or manuscript in place, ask for edits in plain English (remove the gridlines, switch an axis to log scale), and the agent rewrites its own code.
That traceability is not just tidy housekeeping. Under the R&D Tax Incentive, Australian companies claiming the offset must substantiate that the work was genuine, systematic experimental activity. Auditable, reproducible AI workflows are far easier to defend to AusIndustry or the ATO than ad-hoc chatbot sessions with no record of how a result was reached. The same logic applies to grant acquittals and journal submissions, where a reviewer can ask exactly how a figure was generated.
Data stays on your own infrastructure
Claude Science runs on the lab's own machines, so large or sensitive datasets never leave the systems they already sit on, and only the context needed for each step is sent to Claude. For anyone handling health or genomic data under the Privacy Act (medical research institutes especially), that architecture removes one of the standard objections to bringing AI into the workflow. It manages compute for you as well: it drafts a plan, asks before reaching a new resource, and lets you review or revoke any decision before it submits a job to your HPC cluster over SSH or a Modal account, scaling from a single GPU to hundreds as the analysis demands.
What early users are reporting
Anthropic has published three beta case studies. Treat them as vendor examples rather than independent benchmarks, but the pattern across all three is consistent:
Manifold Bio used Claude Science to nominate targets for tissue-targeting medicines end to end, assessing surface expression, trafficking and safety and ranking candidates against criteria drawn from its own proprietary data.
Jerome Lecoq at the Allen Institute built a review pipeline of roughly 20 custom skills. Reviews that once took up to two years now number about 10, many over 100 pages, with citations checked by reviewer agents.
Stephen Francis at the UCSF Brain Tumour Center's glioma epidemiology group ran comprehensive germline workups in roughly one-tenth the time, with the results independently validated by his lab.
The value in each case shows up when a whole multi-step workflow moves into one environment, not when a single query gets answered a bit faster.
Is it worth it for an Australian lab?
The honest answer depends on how much of your current cost is tool-wrangling. Take a mid-sized research group where loaded staff time runs to roughly A$450,000 a year. If a real share of that goes to writing bespoke pipelines, shuffling data between viewers, and manually checking figures against code, a workbench that collapses those steps has a clear business case before you even count faster results. If your work is mostly one trusted tool used well, the case is thinner and you can wait.
Anthropic is also funding early work through an AI for Science program: up to 50 projects with up to US$30,000 in Claude credits each, plus up to US$2,000 in Modal compute for selected projects. Applications are open through 15 July 2026, with projects running September to December. There is a discounted Team plan for academic labs and nonprofit research organisations as well. Note the credits are quoted in US dollars, so budget the exchange rate into any AUD plan.
If you are weighing whether Claude Science fits your R&D setup, or how to keep the resulting workflows defensible for a tax-incentive or grant claim, we help Australian research and product teams make that call. You can book a short session to talk it through.
