Anthropic just made its biggest product move since Claude Code. In the space of one week, the company shipped Claude Sonnet 5, its most agentic Sonnet-tier model yet, and launched Claude Science, a standalone AI workbench built for scientists — with drug discovery as the headline use case. It’s not a small update. It’s Anthropic declaring, in public and in front of a room full of pharma executives, that life sciences is now a core business line, not a side experiment.
This piece breaks down what actually shipped, what it costs, who can access it today, and how it stacks up against OpenAI’s GPT-Rosalind and Google’s Gemini for Science — plus the business context (an approaching IPO, a near-$1 trillion valuation, and a hiring war for AI-for-science talent) that most coverage of this launch has only told in pieces.
Key Highlights at a Glance
- Claude Science launched in beta on June 30, 2026, at a dedicated San Francisco event for pharma executives, biotech founders, and researchers
- It’s available to Pro, Max, Team, and Enterprise subscribers — not a narrow enterprise-only pilot though it currently runs on macOS and Linux only, and Team/Enterprise accounts need an admin to switch it on
- Claude Science bundles 60+ curated skills and connectors across genomics, single-cell research, proteomics, structural biology, and cheminformatics, wired into databases including UniProt, PDB, Ensembl, Reactome, ClinVar, ChEMBL, and GEO
- Anthropic is funding research grants of up to $30,000 to get labs using the platform
- Anthropic is running its own internal drug discovery program aimed at neglected diseases, and demoed the system autonomously identifying candidate compounds for phenylketonuria, a rare genetic disorder
- John Jumper, the Nobel Prize–winning co-creator of AlphaFold, is leaving Google DeepMind after nearly nine years to join Anthropic
- The launch lands alongside a confidential IPO filing with the SEC (filed June 1, 2026) and a Series H round that reportedly values Anthropic near $965 billion
- Claude Sonnet 5 is now the default model in Claude Code and across Free and Pro plans
What Actually Changed
Two releases, two different jobs.
Claude Sonnet 5 is a model upgrade: stronger reasoning, better coding, and more reliable tool use than Sonnet 4.6, with performance Anthropic says approaches Opus 4.8 on several benchmarks at a fraction of the cost. It’s the model most Claude users will notice first, since it’s now the default across Free, Pro, and Claude Code.
Claude Science is a different kind of product entirely — not a model, but a workbench. Anthropic is explicit that the intelligence comes from the underlying model, but the product is about workflow: connecting Claude to lab databases, compute clusters, literature search, and scientific software inside one environment, rather than making researchers hop between a dozen disconnected tools. A scientist can ask it to review recent literature on a target, pull structural data, run an analysis, generate a figure, check the citations, and draft a manuscript section — as one continuous task, with every step traceable back to its source.
That auditability is a deliberate design choice, not a footnote. In science, an answer that looks right isn’t good enough; a result has to be checked against its source data and methodology. Anthropic built Claude Science to keep that trail intact by default.
Why Now: The Business Case Nobody’s Coverage Tells in Full
Most write-ups of this launch tell one of three stories the product story, the talent story, or the money story. All three are actually one story.
The money: Anthropic filed a confidential IPO prospectus with the SEC on June 1, 2026, and closed a Series H round earlier this spring at a post-money valuation near $965 billion. Pharma companies have deeper pockets than academic labs, and Anthropic says it’s approaching its first profitable quarter. A wave of enterprise pharma contracts stemming from Claude Science would matter to that trajectory in a way that consumer chatbot subscriptions never could especially with a public offering reportedly on the horizon later this year.
The acquisitions and partnerships: This launch didn’t come out of nowhere. Anthropic has spent months building toward it acquiring AI biotech startup Coefficient Bio, appointing Novartis CEO Vas Narasimhan to its board, and striking research collaborations with Bristol Myers Squibb and the Gates Foundation on vaccine research, disease modeling, and health-system work in lower-income countries.
The talent war: John Jumper’s move is the loudest signal of the three. He didn’t just work at DeepMind he architected AlphaFold’s deep-learning approach to protein structure prediction, a system that has now modeled more than 200 million proteins and reshaped how drug discovery, structural biology, and materials science get done. He and DeepMind CEO Demis Hassabis shared the 2024 Nobel Prize in Chemistry for that work. Jumper announced his exit on June 19, 2026, after nearly nine years at the lab most responsible for a decade of AI-for-science leadership. Notably, his departure landed the same week Noam Shazeer a key figure behind Google’s Gemini development left Google for OpenAI, underscoring just how aggressively the three labs are now raiding each other’s top research talent.
Put together, the picture is a company fighting on three fronts at once: infrastructure and model costs, commoditizing agent products (which is part of why Sonnet 5 launched cheap and fast), and now a new, high-value scientific-application layer where Google has historically had the strongest claim.
How Claude Science Actually Works
Claude Science is built around a few concrete capabilities:
- Literature and data analysis reviewing papers, comparing targets, and pulling structured data from connected scientific databases in one workflow
- Code execution on real infrastructure Anthropic says the platform can run on a lab’s own compute, from a local machine to an HPC cluster to cloud GPUs, while still surfacing key decisions for a researcher to approve before jobs are submitted
- Domain-specific tooling more than 60 skills and connectors covering genomics, single-cell analysis, proteomics, structural biology, and cheminformatics, linked to databases like UniProt, PDB, Ensembl, Reactome, ClinVar, ChEMBL, and GEO
- Reproducibility by default every figure or result carries a traceable record back to its source data and the code that generated it
In beta, Anthropic says researchers have already used the platform for single-cell RNA sequencing analysis, CRISPR screen design, protein structure prediction, and cheminformatics work early-stage research tasks rather than anything resembling a finished, regulator-ready drug candidate.
Availability and Pricing
Claude Science is in beta and currently limited to Pro, Max, Team, and Enterprise subscribers, running on macOS and Linux (no Windows support yet). Team and Enterprise accounts need an admin to enable it before individual users can access it. Anthropic is also offering research grants of up to $30,000 to encourage academic and nonprofit labs to adopt the platform early — a detail that matters for university researchers weighing the cost of switching tooling mid-project.
Claude Science and Drug Discovery
The headline use case is drug discovery, and Anthropic is putting its own resources behind it. The company announced it will run an internal drug discovery program using Claude Science, focused specifically on neglected diseases conditions with real disease burden that traditional pharmaceutical and biotech investment tends to skip because they aren’t commercially attractive.
Anthropic’s life sciences head, Eric Kauderer-Abrams, has framed the internal program as a feedback mechanism as much as a humanitarian one: building AI tools for an industry Anthropic hasn’t actually worked inside of is harder than building tools alongside real, firsthand development experience. Jonah Cool, Anthropic’s head of life sciences partnerships, has described the neglected-disease focus as running in parallel with the company’s broader push to sell Claude Science to biopharma customers using the internal research as a credibility-building exercise with the exact companies Anthropic is trying to court.
At the launch event, Claude Science’s development lead demonstrated the system autonomously identifying drug candidates for phenylketonuria, a rare genetic disease a controlled, concrete proof point rather than a claim about general drug discovery capability. Anthropic hasn’t said what happens if the internal program surfaces a genuinely promising compound; in the standard pharma pipeline, that would mean moving into costly, multi-year clinical trials, a step well beyond what any AI tool can do on its own.
Claude Science vs. GPT-Rosalind vs. Gemini for Science
Claude Science enters an already crowded field. All three major labs now have a dedicated science product, and each has taken a different shape.
| Claude Science (Anthropic) | GPT-Rosalind (OpenAI) | Gemini for Science (Google) | |
|---|---|---|---|
| Format | Standalone workbench, same product tier as Claude Code/Cowork | Purpose-built model, paired with Prism (writing/collaboration workspace) | Suite bundling Hypothesis Generation (built on DeepMind’s Co-Scientist), Literature Insights, Science Skills |
| Core strength | Execution: running code, managing compute, reproducibility | Specialized reasoning in biology, medicinal chemistry, genomics | Deep bench of proprietary scientific models, including AlphaFold |
| Database integration | 60+ connectors (UniProt, PDB, Ensembl, Reactome, ClinVar, ChEMBL, GEO) | Plugin-based database access | 30+ life-science databases wired directly into the suite |
| Headline demo | Autonomous drug candidate identification for phenylketonuria | Research and drug discovery acceleration (launched April 2026) | Hypothesis generation via Co-Scientist |
| Talent signal | Hired John Jumper (AlphaFold co-creator) from DeepMind | Building a fully autonomous AI researcher (stated goal) | Nobel Prize pedigree via Hassabis and Jumper’s original AlphaFold work |
The real differentiator isn’t whether each company has a science product they all do now. It’s how each one is packaged. Claude Science leans into hands-on execution: writing and running code, managing cluster jobs, keeping a reproducibility trail. OpenAI splits the job between a specialized reasoning model and a separate collaboration workspace. Google leans on its AlphaFold-era database depth and its head start in the category.
For drug discovery specifically, this is genuinely a three-way race rather than a category Claude has to itself Jumper’s hire is best read as Anthropic trying to close a talent and credibility gap on Google’s home turf, not as evidence it’s already ahead.
Claude Sonnet 5: The Quieter Half of the Update
Sonnet 5 doesn’t come with a launch event of its own, but it’s the part of this update that touches the most users day to day. It’s built to hold onto multi-step problems from start to finish rather than losing the thread halfway through, and it’s now the default model in Claude Code, in Free, and in Pro plans. For research teams, it matters beyond general coding: a large share of modern scientific work runs through code that most scientists were never formally trained to write, which is part of why coding-capable models have already become common in labs — and part of why Sonnet 5 functions as the practical engine underneath Claude Science’s workflows.
Independent validation has come from outside Anthropic, too. Harvard physicist Matthew Schwartz, describing his own use of Claude Code and related Anthropic tools, put Claude Opus 4.5’s ability to execute scientific project work at roughly the level of a second-year graduate student. It’s a useful, if imperfect, anchor: a second-year graduate student still needs oversight from a principal investigator, and Anthropic’s own framing treats Claude Science the same way — a highly capable research assistant, not an unsupervised scientist.
Who Should Actually Use This
Developers get a sharper, more dependable Claude Code by default no separate signup needed.
Researchers in molecular and cellular biology, genetics, chemistry, and drug discovery are the core audience for Claude Science, with the tooling built specifically around their databases and workflows.
Biotech and pharma companies are, literally, who Anthropic built this for the launch event was staged for that exact audience, and the commercial logic (deep-pocketed customers, an approaching IPO) is not subtle.
Academic and nonprofit labs can access Claude Science on Pro, Max, Team, or Enterprise plans, and the $30,000 research grants specifically target this group, even though the product is clearly built with professional pharma research as the priority.
Students get more value from Sonnet 5’s general reasoning and writing help than from Claude Science itself, unless they’re doing hands-on lab or computational research.
Clinicians and healthcare administrators are not really the target here Claude Science is aimed at the research and drug-development side of the pipeline, not clinical or administrative healthcare work.
Limitations Worth Knowing
- Access isn’t fully self-serve for larger teams. Team and Enterprise users need an admin to enable Claude Science before anyone on the account can use it.
- Every output still needs expert review. Reproducibility features make it easier to check an answer, but they don’t replace scientific judgment especially in biology and drug discovery, where a wrong assumption can be costly.
- “Autonomous” has boundaries. The phenylketonuria demo was a candidate-identification exercise, not a finished, trial-ready compound. Anthropic hasn’t outlined what its internal program would do if it did find something promising that path still runs through the normal, years-long clinical trial and regulatory process.
Sonnet 5 is the everyday upgrade faster, more reliable, now the default almost everywhere Claude shows up. Claude Science is the strategic bet: a standalone product, backed by a Nobel laureate hire, a growing list of pharma partnerships, and a near-trillion-dollar valuation heading into an IPO, aimed squarely at an industry with far more money to spend than the consumer chatbot market ever offered. Whether Claude Science becomes a genuine fixture in drug discovery pipelines rather than a well-marketed research assistant is the question the next year of pharma contracts, and Anthropic’s own internal drug program, will actually answer.
