On Tuesday June 30, 2026, Anthropic launched Claude Science — an AI research workbench with more than sixty curated skills and connectors covering genomics, single-cell, proteomics, structural biology, and cheminformatics — in beta to every paid Claude subscriber. Pro, Max, Team, and Enterprise plans all get the same workbench, per Anthropic's own launch note and Drug Discovery World's write-up on July 6. The Pro tier starts at $20 a month. That is the number to hold in your head.

That is because the shape of the announcement is not "we shipped a new science model." The shape is "we shipped a science workbench to every subscriber on the paid tier the trade press hardly writes about." OpenAI's Life Sciences 36-hour stack shipped two weeks ago inside a named-pharma-partner motion — Novo Nordisk, Eli Lilly, Moderna, Sanofi. Anthropic just answered that with the individual-researcher motion. Different audience, different sales cycle, same beat.

What sixty tools actually looks like

Per Anthropic's launch and Pharmaceutical Technology's coverage, the workbench is built around a generalist coordinating agent that dispatches to specialist skills and to external connectors. The connectors named across the two write-ups include PubMed, Jupyter, R, plus NVIDIA's BioNeMo Agent Toolkit and Modal for on-demand compute. The specialist skills sit under the same generalist Claude the API already ships, so the interaction pattern is the one every Claude user already knows: type a question in plain language, and the agent decides which of the sixty tools to reach for.

The concrete tasks named in the industry write-ups are the ones that give the sixty-tool number weight. CRISPR screen design. Single-cell RNA sequencing analysis. 3D protein structure rendering. Molecular epidemiology analyses. Iteratively refining figures and manuscripts until they are ready for publication. Those are five different domains with five different data-file formats and five different tool stacks a researcher would otherwise have to install by hand. That is the reason a coordinating-agent-plus-skills architecture is the right shape for this problem: no lab wants to install five separate copilots.

The Pharmaphorum write-up surfaces the single most useful reader-facing anecdote. One researcher completed a molecular epidemiology analysis "in around a tenth of the time it previously took." A single-datapoint testimonial is not a benchmark and Anthropic does not pretend it is, but the shape of the number — an order of magnitude, not a percentage — is consistent with what the "one workbench" pitch is actually selling. The gain is not accuracy; it is not context-switching between six terminals.

The Pro-plan door

The most quietly load-bearing detail in the launch is that Anthropic did not gate Claude Science behind an Enterprise contract. The beta ships on Pro, Max, Team, and Enterprise, per Anthropic's own note. Pro is $20 a month. Max is $100 or $200 a month depending on the seat count. Neither requires a procurement call, a legal review, or a named account executive.

That door being open is a real shift. OpenAI's Life Sciences push is enterprise-partnership shaped; Google's Gemini life-sciences tooling, per Pharmaceutical Technology, sits inside its healthcare-and-life-sciences vertical. Anthropic has effectively said: if you have a Pro subscription and a PhD, the workbench is yours today. That inverts the usual pharma-AI sales motion, which is close the enterprise deal, then figure out who inside the enterprise actually uses it.

The under-the-line reading is that Anthropic thinks the individual researcher is the wedge into the enterprise account, and that giving the postdoc a workbench today is worth more than making her wait for her institution's IT policy to catch up. That is a very Claude-Code-shaped bet. It is the same reason Cursor and Claude Code got there before the sanctioned enterprise-AI-IDE procurement cycles finished.

The grants program is the flywheel

Alongside the launch, Anthropic opened an AI for Science grants program. Up to fifty projects, up to $30,000 in Anthropic credits per project, with Modal kicking in up to $2,000 in compute credits for select recipients, per Drug Discovery World and Pharmaphorum. Applications close July 15. Awards go out July 31. Projects run September 1 through December 1, 2026.

Fifty times thirty thousand is one and a half million dollars in inference credits, which is not a large cheque by pharma standards and is not the point. The grant is the recruiting funnel. The point is to seed fifty published case studies in the fall — one per lab — that Anthropic can point at when the enterprise sales team walks into Genentech in January. It is the exact same shape as the Claude Corps fellowship program from June, translated from software engineers to scientists, at roughly one hundredth the budget because a grant of thirty thousand dollars in credits is much cheaper to underwrite than a stipend and a salary.

The applications window is nine business days. The awards window is another two weeks. That is a short fuse for a scientific grant program, and it is short on purpose. Anthropic wants the fifty projects running before OpenAI or Google announce their equivalent.

The Bristol Myers Squibb footprint already booked

Per Pharmaceutical Technology, Bristol Myers Squibb rolled out Claude to more than 30,000 employees in May 2026 — before Claude Science existed as a named product. That is the customer already sitting inside the account when the workbench shipped, and it is the reason Anthropic can credibly say the Pro-tier motion is a wedge and not a giveaway. The individual-researcher door is open on Pro; the enterprise door is already ajar at one of the ten largest pharma companies on the planet.

Anthropic also confirmed, per DDW, an internal drug-discovery program targeting neglected diseases. The specifics are thin — no molecule name, no disease list, no timeline — and until they land the program sits somewhere between research initiative and press-release colour. But the shape is worth flagging. A lab that ships an external science workbench and simultaneously runs its own drug program is, on paper, both a supplier and a competitor to its own pharma customers. That works right up until it doesn't. Ask any cloud vendor that has ever built its own SaaS on top of its own infrastructure.

The competitive frame is now three-lane

Reading the three current life-sciences AI plays side by side:

OpenAI Life Sciences (June 21). Named pharma partners — Novo Nordisk, Eli Lilly, Moderna, Sanofi. The 36-hour stack, LifeSciBench, the Rosalind evaluation. Enterprise-first, benchmark-and-partner shaped. High-touch sales motion.

Google Gemini for Life Sciences. Sits inside Google Cloud's healthcare-and-life-sciences vertical, per the Pharma Tech write-up. Hyperscaler-attached. Sells alongside Vertex AI and BigQuery.

Anthropic Claude Science (June 30). Sixty tools, four subscription tiers including Pro, an internal drug-discovery program, and a fifty-project grants funnel. Individual-researcher-first, workbench shaped. Low-friction motion aiming to be the tool the postdoc already uses when the enterprise procurement call finally happens.

The three motions are not exclusive and the three companies will each hire the sales teams to sell all three. But the shape of the first launch tells you what the company thinks its edge is. OpenAI thinks its edge is the pharma-partner Rolodex it built in 2025. Google thinks its edge is the cloud contract. Anthropic thinks its edge is the individual scientist typing a question into a Claude window on Tuesday afternoon.

What this means

1. The Pro-tier door is the story. "Sixty tools on a twenty-dollar subscription" is a sentence a pharma buyer's IT department will not process the same way it processes an OpenAI enterprise contract. It is also a sentence individual researchers will process very quickly. Anthropic did not accidentally make Claude Science available on Pro. That is the wedge.

2. The grants program is a recruiting funnel with a nine-business-day window. Fifty projects, thirty thousand each, applications closing July 15 and awards out by July 31. That is the timeline of a company that wants a case-study inventory built before its rivals ship equivalents. If you are a life-sciences PI reading this, the application deadline is a week from Wednesday.

3. The workbench pattern is now the standard shape. OpenAI's Life Sciences push was a stack. Anthropic's is a workbench. The Bridgewater / Thinking Machines fine-tune paper from June 30 is a training pipeline. Three different labs, three different weeks, three different verticals, three variations on don't sell the model, sell the environment around the model. That is the industry's answer, for now, to the question of what to sell when the model itself is a commodity.

4. Bristol Myers Squibb is the tell. A 30,000-seat enterprise deployment predating the science-workbench launch tells you the Pro-tier motion is not a giveaway strategy. It is a two-lane motion. Individual researchers get the workbench on their own subscriptions; the enterprise account team walks in later with an existing footprint to expand. That is the Claude Code playbook applied to pharma.

Anthropic did not announce a partnership with a hedge fund, a data-centre lease, or a chip. It shipped a research tool to every paid subscriber and opened a grants program with a July 15 deadline. In the week the Loop covered a $30 billion analyst-labelled fine-tune and a nine-percent Meta rumour, that was the announcement most likely to be quietly running on ten thousand PhD laptops by August.