Anthropic PBC on Tuesday put Claude Science, an AI workbench aimed at bench scientists, into beta for its paying customers, weeks before its expected fall initial public offering and with two of the drug industry's largest chief executives onstage to vouch for the use case.
The launch, hosted in San Francisco with CEO Dario Amodei joined by Bristol-Myers Squibb chief executive Chris Boerner and Novartis chief executive Vas Narasimhan, is the clearest signal yet that Anthropic is selling its IPO story on enterprise toolchains, not on another round of frontier-model scaling. Financial Post reports Anthropic's release of Claude Science as automation software for research workflows, and MIT Technology Review frames it as Anthropic's newest flagship product, not a model release. TechCrunch's read is the same: workflow over new model.
What Claude Science actually does, per Anthropic's announcement and the mer.vin explainer of the workbench, is bundle more than 60 scientific databases behind a plain-language interface so a researcher can run multistep tasks, including protein-structure prediction, without stitching together separate notebooks, databases, and tool calls. Critically, Anthropic is not introducing a new model: Dataconomy's launch writeup notes the product runs on existing Claude models, including Opus 4.8 released in May 2026, with output metadata that lets scientists trace what data and steps produced a given result. First Post's coverage of the launch highlights the same audit-friendly framing.
The strategic context is the IPO math. Anthropic is now valued at roughly $965 billion and is reported to be pushing for a fall 2026 listing. That timing matters because, five months earlier, Anthropic shipped a different enterprise wedge: Claude Cowork, a legal-work automation product that, per the same Financial Post account, helped trigger a roughly $1 trillion drop in equity values tied to potentially disrupted industries. Investors are no longer pricing each Anthropic launch as a science update; they are pricing it as a discovery event for the next industry in the crosshairs.
The beta's audience helps explain why Wall Street is watching. Anthropic's announcement names Bristol-Myers Squibb and Novartis as launch partners and lists Boerner and Narasimhan as speakers. Narasimhan is also on Anthropic's board, per Global Healthcare Magazine's coverage of the launch, so the relationship predates the product, but a joint stage appearance is the kind of evidence public-market investors will look for when weighing enterprise traction. Neither company has publicly confirmed a commercial deployment of Claude Science; pharma CEO appearances at a launch event are a design-partner signal, not a procurement commitment.
Anthropic is also publishing the credibility scaffolding. Alongside Claude Science, the company released BioMysteryBench, a benchmark it built to evaluate Claude on bioinformatics tasks, with Anthropic claiming parity with human experts. The Decoder flags that this is a vendor-run benchmark, which means independent evaluation is needed before the claim can be treated as settled. Read alongside the workbench itself, the move is a paired punch: ship a tool that biologists can use, then ship a test that says the underlying model is good at biology.
The reader-side value is more grounded than the marketing. Bench scientists spend hours pulling sequences, querying databases, and copying outputs between tools, and mer.vin frames Claude Science as automation of those tedious steps rather than autonomous science. Whether the product survives contact with regulated, IP-heavy pharma workflows is the next thing to watch. The single-sourced February Cowork precedent, the absence of disclosed enterprise pricing, and the still-narrow paid-customer beta rollout mean revenue impact is, for now, a story about expectations, not earnings.