Google's ERA Tool Reaches Nature, Expands to Google Labs
Google's Empirical Research Assistance tool, which uses Gemini to write and optimize scientific code, is described in a new Nature paper and is now rolling out through Google Labs.
What is Google's ERA tool and what does it actually do?
Empirical Research Assistance (ERA) is a Google-built AI tool that writes and optimizes code for computational science experiments. Given a scientific problem and a success criterion, ERA searches literature, generates code, explores candidate solutions, and evaluates results using a tree-search approach that considers thousands of options.
Nature publication and benchmark results
A paper describing ERA's design and performance appeared in the journal Nature on May 19, 2026, authored by Google Research. The paper covers benchmark testing across genomics, public health, satellite imagery analysis, neuroscience prediction, time-series forecasting, and mathematics. According to the publication, ERA achieves expert-level performance across all of those benchmarks.
The benchmarks span meaningfully different scientific domains, which is notable because it suggests the tool is not narrowly tuned to one type of problem. The Google team frames this breadth as potentially expanding access to expert-level computational modeling for researchers who lack deep domain-specific coding expertise.
From preprint to product
Google first released ERA as a preprint last fall, so the Nature publication formalizes work that had already been circulating in the research community. Over the intervening months, Google Research scientists and outside collaborators ran ERA on open scientific questions, producing a total of eight manuscripts. Five of those papers were released alongside the Nature publication.
One example cited in the source is epidemiological forecasting: researchers used ERA to predict U.S. hospital admissions at the state level up to four weeks ahead. The breadth of the applied work, spanning at least eight separate scientific problems, gives a clearer picture of where the tool performs in practice versus controlled benchmarks.
Computational Discovery and Google Labs access
ERA is one of the systems underpinning Computational Discovery, described as a new experimental tool. As of May 19, 2026, Computational Discovery is beginning a broader rollout through a trusted tester program under Gemini for Science in Google Labs. The announcement was made as part of Google's I/O event.
The relationship between ERA and Computational Discovery matters for understanding what is actually shipping. ERA is the underlying AI coding system; Computational Discovery is the prototype product built on top of it. Researchers interested in access would go through Google Labs rather than directly to the ERA codebase, though Google also published ERA's code and experiments alongside the Nature paper.
What the tool addresses
The core problem ERA targets is the iterative nature of computational experiments: writing code, running it, interpreting results, adjusting, and repeating. That cycle is a significant time sink in empirical research, and automating it is the tool's stated purpose. ERA does not replace the scientist's judgment about what question to ask or what success looks like; it requires both as inputs.
Google has released the ERA code and accompanying experiments publicly, which allows independent researchers to examine the approach and reproduce the benchmark results described in the Nature paper.