Anthropic just launched Claude Science, its newest flagship product, and the company is placing it on the same level as Claude Code and Claude Cowork. According to MIT Tech Review, Eric Kauderer-Abrams, Anthropic’s head of life sciences, calls it a core piece of the company’s mission. “It represents how important this is to our mission that this is right up there with Claude Code and Claude Cowork as the next really significant product that we’re releasing,” he told MIT Tech Review. “Our mission is to develop AI that serves humanity’s long-term well-being, and we believe that by far the greatest opportunity to do that is in the life sciences.”
What stands out here is the timing. For a decade, Google DeepMind owned AI for science, with Demis Hassabis and John Jumper winning the Nobel Prize in chemistry for AlphaFold. Now Anthropic is moving onto that turf, and it just landed a major vote of confidence: earlier this month, Jumper announced he’s leaving DeepMind for Anthropic.
What Claude Science actually does
Kauderer-Abrams was clear that this isn’t meant to replace tools scientists already lean on. It builds on them. Based on the MIT Tech Review report, here’s what sets it apart:
- It writes code, but goes further. A lot of modern research involves coding, yet most scientists aren’t trained software engineers. Claude Science picks up where tools like Claude Code left off.
- It runs your code on big clusters. Many scientists need powerful computer clusters, and managing them is a headache. Claude Science helps handle that part directly.
- It prioritizes reproducibility. The tool is built so researchers can trace any figure or result back to its source and check it for accuracy and validity. That’s a big deal in a field where a single unverifiable result can sink a paper.
Why Anthropic thinks it can win
MIT Tech Review points to a few reasons Anthropic is well positioned to take DeepMind’s scientific mantle. CEO Dario Amodei is a PhD scientist, much like Hassabis, and unlike OpenAI’s Sam Altman, who came up as a businessman. Scientists are already heavy users of Claude Code, so the workflow habit is there.
There’s also the raw capability jump. Since LLM agents became genuinely useful for independent work in late 2025, researchers have been testing their limits. In a blog post on Anthropic’s site, Harvard physicist Matthew Schwartz estimated that the company’s Opus 4.5 model is roughly as capable of executing scientific projects as a second-year graduate student. That’s a striking benchmark, and it explains why Anthropic sees life sciences as its biggest opportunity.
Why this matters
This is significant because the AI-for-science race just got a real second player. MIT Tech Review notes that DeepMind, long the leader, has been left playing catch-up on coding, which has become the most lucrative use case for LLMs. Anthropic is attacking from exactly that strength. If the most valuable skill in AI right now is writing and running code, and a huge chunk of modern research depends on code, then a science-focused product is a natural extension rather than a side bet.
The reproducibility focus is the piece I’d watch most closely. Speed and code generation are useful, but science lives or dies on whether results can be checked. A tool that keeps a clean trail from raw data to final figure could earn trust faster than one that just produces answers quickly.
A few things stay open. The article doesn’t spell out pricing, availability, or which disciplines get first access, so the practical rollout details are still to come. What’s clear is the ambition: Anthropic isn’t treating science as a niche vertical. It’s treating it as a flagship.
For the full breakdown and the quotes from Anthropic’s team, the original report at MIT Tech Review is worth a read.