Claude Science: Anthropic positions itself as the new DeepMind and takes direct aim at the pharmaceutical industry

By Zendoric · June 30, 2026.

For years, the name Google DeepMind was synonymous with AI for science. AlphaFold, the Nobel Prize in Chemistry shared by Demis Hassabis and John Jumper, contributions to meteorology and materials science... DeepMind built a reputation that no competitor seemed capable of seriously threatening. That changed this week, and Anthropic is what changed it.

At a closed-door event for pharmaceutical executives, biotech founders and researchers, the company unveiled Claude Science: an autonomous scientific research product designed to do for biology and drug development what Claude Code does for software engineering. It is not an extension or a plug-in —in October 2025 a more limited version already existed under the 'Claude for Life Sciences' umbrella—, but a top-tier product within Anthropic's lineup, on the same level as Claude Code and Claude Cowork. It is already available to all paying subscribers.

The launch includes concrete capabilities: autonomous execution of analysis pipelines on compute clusters, integration with genomics, chemistry and protein biology tools, and an explicit emphasis on reproducibility —so that any result or figure can be traced back to its source and audited. During the presentation, Alexander Tarashansky, who led the product's development, demonstrated how the system autonomously identified drug candidates for phenylketonuria, a rare genetic disease. And Anthropic is not leaving that work solely to third parties: the company has announced it will use Claude Science to drive its own research into neglected diseases.

**Zendoric's thesis: this is not just a product, it's a changing of the guard.**

The combination of three factors makes this launch something qualitatively different from the sector's usual noise. First, Dario Amodei himself holds a doctorate in science —he is not a businessman who talks about science for marketing purposes, as the article notes in comparing him with Sam Altman—. Second, a significant fraction of Claude Code's most active users were already scientists: people who need to write code to analyze data but are not engineers by training. Claude Science turns that ad hoc use into an experience designed specifically for them. And third —the most symbolic blow—, John Jumper, co-author of the work that earned DeepMind the Nobel, has just signed with Anthropic. When the lead architect of your most celebrated achievement leaves for the competition, the narrative shifts.

There is also a financial component it would be naive to ignore. The article states it clearly: the pharmaceutical industry has far deeper pockets than academia. Anthropic is about to close its first profitable quarter, and is preparing for an IPO before the end of 2026. In that context, steering its flagship science product toward drug discovery —one of the few sectors that can pay for AI tools at scale— is corporate strategy as legitimate as it is necessary. The humanitarian mission and the business model coincide here in an unusually comfortable way, and that deserves to be celebrated with caution: the incentives are reasonably aligned, but incentives can always change.

As for the product's real capability, the only public benchmark the article cites is from Harvard physicist Matthew Schwartz, who estimated that Anthropic's Opus 4.5 model has a scientific execution capability comparable to that of a second-year doctoral student. It's a useful and honest figure: the claim is not that AI cures diseases on its own, but that it accelerates the work of those who research them. That is the correct framing. A second-year student doesn't discover drugs; they help identify them, filter candidates, run computational experiments that would have taken weeks. Multiplied across hundreds of simultaneous researchers, the potential impact is real even if still hard to quantify.

**What this means for DeepMind —and for the sector.**

As sector context, Google DeepMind remains a formidable player: its Gemini models compete at the general frontier, and its legacy in science is undeniable. But the article points to something we have been observing for months: in the most lucrative use case of the moment —autonomous programming—, DeepMind is arriving late. Now Anthropic is occupying the AI-for-science space with a dedicated product, with talent recruited directly from DeepMind, and with the strategy of going first to where the money is most abundant: the pharmaceutical industry. DeepMind will have to respond, and the response will take time.

The question that remains open —and that will define Claude Science's real success over the next two years— is whether the product will be genuinely accessible to academic science, which is where the most disruptive discoveries tend to happen but where budgets are tightest. If Claude Science ends up in practice reserved for pharmaceutical companies with enterprise contracts, the humanitarian promise becomes a marketing argument. If Anthropic keeps prices affordable or strikes institutional agreements with universities, the story is different.

Either way, there is something in this move that connects directly with the long-term thesis we maintain here: AI applied to molecular biology and drug discovery is not speculative hype. It is one of the most concrete and verifiable indicators that the coming decades may bring something genuinely transformative in human health. The transition will be uneven and access will not be guaranteed for everyone from day one —that's the hard part—. But the fact that a private company is investing its best resources in turning scientific research into an automatable and reproducible process is, objectively, one of the best pieces of news of the year.