Author(s): Chew Loong Nian – AI ENGINEER

Originally published on Towards AI.

Anthropic Crushed a 2-Year Science Job Into Days — With Zero New Models

A neuroscientist at the Allen Institute used to spend up to two years writing a single long-form computational review. Since late June he has produced about ten of them, many running past 100 pages, each with citations that a second AI checked line by line. The tool that did it is Anthropic’s biggest science bet of the year — and here is the part that broke my brain: it contains no new model.

After the introduction, the article explains how Claude Science turns a workflow that normally takes years into days by using a computational review template with multiple agents, curated skills/connectors, and the ability to generate scientific artifacts and figures. Its key innovation is an actor–critic setup: a coordinating “creator” agent drafts the work while a separate reviewer agent fact-checks citations, numbers, and figure/code consistency step-by-step to reduce hallucinations and improve auditability. The release is also framed as a reproducibility tool, since each generated figure ships with the exact code, environment, methodology description, and full message history so results can be regenerated and sessions forked. The author compares Anthropic’s approach with OpenAI’s GPT-Rosalind and Google’s Gemini-for-Science stack, argues that workflow and auditing are the real bottlenecks rather than a brand-new model, and provides quick-start guidance (desktop app or running via SSH/HPC). Finally, it covers caveats—beta stage, lack of public head-to-head benchmarks, reviewer agents reduce but don’t eliminate errors, outputs still need wet-lab validation, and users are more tightly tied to the Claude ecosystem—before concluding that Claude Science is especially valuable for academic labs and researchers who want reproducible results with minimal extra setup.

Read the full blog for free on Medium.

Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor.

Published via Towards AI