By Anthropic shipped Claude Science on June 30 — a desktop AI workbench that replaces the fragmented research stack scientists have tolerated for decades. It pulls 60+ scientific databases, multi-agent pipelines, and compute management into one environment. It runs on Claude Opus 4.8, costs nothing extra for paid subscribers, and is already producing results worth paying attention to: a UCSF researcher found viral contamination in RNA-seq data within minutes — contamination that had gone undetected for a full year using conventional methods.
One Environment for the Entire Research Pipeline
The traditional computational biology workflow is a stitched-together mess: PubMed for literature, Jupyter for code, R for statistics, a separate HPC terminal for cluster jobs, and five or six databases — each with its own schema and query language — that you have to translate between by hand. Claude Science replaces that workflow with a single environment you direct in plain English.
Under the hood, a generalist coordinating agent delegates work to domain-specific specialist agents. Those agents query databases like UniProt, PDB, Ensembl, Reactome, ClinVar, ChEMBL, and GEO — 60+ in total — write and execute code on your lab’s HPC cluster via SSH or scale to on-demand cloud GPUs through Modal, and natively render 3D protein structures, genome browser tracks, and chemical structures without exporting to another tool. The compute layer is the key detail: it asks for your approval before spinning up new resources, and scales from a single GPU to hundreds as needed.
NVIDIA’s BioNeMo Agent Toolkit is integrated directly, giving Claude Science access to Evo 2 (a DNA foundation model), Boltz-2 (biomolecular interaction prediction), and OpenFold3 (protein structure prediction) as callable skills. These are not wrappers — they are BioNeMo NIM microservices packaged as containerized inference endpoints, pre-tuned and ready to run.
The Reproducibility Differentiator
Every output Claude Science generates includes an auditable history: the exact code that produced it, the environment it ran in, a plain-language description of what happened, and the full conversation that led to it. When it generates a figure, you get everything you need to reproduce that figure from scratch. A dedicated reviewer agent runs in parallel throughout, flagging incorrect citations, untraceable numbers, and figures that do not match their underlying data.
This is not a convenience feature. Scientific irreproducibility is a genuine crisis — a substantial fraction of published computational biology results cannot be independently reproduced, largely because researchers cannot reconstruct the exact conditions under which an analysis was run. Claude Science treats reproducibility as a hard requirement, not an afterthought, and that makes it a fundamentally different class of tool from what came before.
Early Evidence It Works
Claude Science immediately found a laboratory virus contaminant in our bulk RNA-seq data. We spun our wheels on this for the better part of a year, and it came out as one of the first key findings.
Prasad Shirvalkar, Associate Professor of Neurosurgery and Anesthesiology, UCSF
In a separate case, Claude Science analyzed 100 rare genetic diseases in under an hour and flagged 32 candidates for computational screening — a task that would typically require weeks of specialist effort.
How to Access Claude Science Now
Claude Science is available today in beta on macOS and Linux for Claude Pro, Max, Team, and Enterprise subscribers. There is no separate waitlist and no enterprise gate. It runs on Claude Opus 4.8 — the same model already in your subscription — so there is no additional cost.
Alongside the product launch, Anthropic opened its AI for Science grant program: up to $30,000 in Claude Science credits and dedicated compute for 50 selected research projects. Applications close July 15. Award notifications go out July 31, and projects run September 1 through December 1, 2026.
The Drug Discovery Angle
Claude Science did not launch in a vacuum. Anthropic simultaneously announced it is running its own in-house preclinical drug discovery programs, targeting neglected diseases that Big Pharma considers unprofitable. The rationale is straightforward: real drug discovery experience creates a feedback loop that makes Claude Science better, which attracts biopharma customers who pay for it. According to CNBC, Claude Science can already handle single-cell RNA sequencing and CRISPR design workflows autonomously — tasks that would normally require a team of specialists and weeks of effort.
The NVIDIA BioNeMo partnership extends this further, giving researchers access to life sciences foundation models through the BioNeMo Agent Toolkit without having to stand up the infrastructure themselves. MIT Technology Review called Claude Science Anthropic’s newest flagship product — and that framing is hard to argue with.
If you are a researcher or developer working in computational biology or building research tooling, Claude Science is worth evaluating today. The access barrier is low — Pro subscription, macOS or Linux — and the July 15 grant deadline is close enough to act on now.
