Anthropic Launches Claude Science, Betting on Research Workflows
Taylor Wilson
Anthropic on June 30 launched Claude Science, an AI research workbench for scientists that bundles databases, coding tools and compute into a single environment; the move signals the $900-billion-valued company is shifting its growth bet from model superiority to vertical workflow products.
What exactly is this product?
Claude Science is not a new model. It runs on existing Claude models — including Claude Opus 4.8 — with no special access or feature gates.
Its differentiator is workflow integration: 60-plus scientific databases, pre-built toolkits for genomics, protein structure and chemistry, a coding environment and compute — all in one interface.
In plain terms = researchers used to juggle a dozen separate tools; Anthropic wants them to run an entire computational study without leaving the workbench.
How does it actually work?
At the core is a lead AI assistant acting as a "project manager." It can spawn sub-assistants for parallel tasks or hand work off to user-built "expert" agents — this means → it is a multi-agent orchestration system, not a single chat window.
A separate fact-checking AI verifies citations and calculations before results are published — but note: the checker runs on the same underlying model, not an independent external source.
For reproducibility, every scientific visualization — 3D protein structures, genome-browser tracks, chemical diagrams — ships with the code and runtime that generated it, plus a full message history. Data can also stay on a lab's own infrastructure, with nothing uploaded to Anthropic's servers.
Is anyone actually using it?
Sean Whalen, chief scientist for machine learning at the Gladstone Institutes, built a genome browser from scratch in days using Claude Science.
Jérôme Lecoq, a neuroscientist at the Allen Institute, built a multi-agent computational review pipeline that reportedly saved years of manual work.
Both cases were disclosed by Anthropic itself — early-user endorsements with no independent third-party verification yet.
What is the business case?
Claude Science is available in beta to all Pro, Max, Team and Enterprise subscribers — this means → it is not a standalone paid product but a lever to drive subscription upgrades.
Anthropic has signed partnerships with Novo Nordisk and AstraZeneca, and in April acquired AI drug-discovery startup Coefficient Bio.
This reflects a clear pre-IPO revenue playbook: lock in pharma enterprise clients with workflow tools → then use that enterprise revenue to support the $900 billion valuation narrative at listing.
How do the three giants' strategies diverge?
Anthropic bets on workflow integration: no bespoke science model — wrap existing models into a vertical-industry workbench.
OpenAI bets on a dedicated model: launched GPT-Rosalind in April, fine-tuned for biology reasoning, available only after qualification review. Partners include Amgen, Moderna and Thermo Fisher.
Google DeepMind bets on foundational-model ownership: it owns AlphaFold and AlphaGenome outright — the other two can only call them as external tools.
In plain terms = the core split is "build your own engine" versus "build the car around someone else's engine." Pharma procurement decisions will be the scoreboard.
Any other details worth noting?
Anthropic will fund up to 50 Claude Science projects, each with up to $30,000 in compute credits, open to postdocs and graduate students.
Application deadline is July 15, 2026; project period runs September 1 to December 1.
This means → Anthropic is spending real money to seed academic-user habits — a classic developer-ecosystem play, penetrating from academia toward industry.
Content is for reference only, not financial advice.