← About ESM3
ESM3 OVERVIEW

Did ESM3 Simulate 500 Million Years of Evolution?

An evidence-led explanation of the ESM3 fluorescent protein result and the 500-million-year comparison in the original Science publication.

In brief: The “500 million years” phrase is an evolutionary-distance analogy derived from the research, not a claim that ESM3 literally simulated five hundred million years of ecological and genetic history.

What the original research reported

In the Science paper, researchers demonstrated generative protein design and described a functional fluorescent protein that differed substantially from naturally occurring known relatives. This was a genuine experimental result within a broader study, not proof of general-purpose artificial evolution.

Understanding the number

The stated evolutionary timescale was inferred by comparison with natural evolutionary changes and sequence divergence. Such an estimate depends on assumptions and the relevant comparison set. It should not be read as a precise stopwatch measurement or evidence that a single generated protein captures all historical pathways.

Why the distinction matters

The result illustrates how a generative model may reach distant points in protein sequence space, but it does not establish that every model proposal will fold or function. Reproducibility depends on prompt design, sampling procedures, model checkpoint and subsequent laboratory experiments.

How to cite the result

Use the peer-reviewed article: Hayes et al., “Simulating 500 million years of evolution with a language model,” Science (2025), DOI: 10.1126/science.ads0018. Distinguish the initial 2024 announcement from the journal publication.

Primary sources and further reading

Editorial review: October 2026. This guide is educational. Model availability and APIs may change; confirm exact versions in the linked documentation. Generated structures and sequences require independent scientific validation.

Scientific context

Research education only. Outputs from protein AI models are predictions and require independent evaluation. Verify APIs, model availability and licensing with official documentation.

✦ AI RESEARCH ASSISTANT

Questions about this article?

Get a concise explanation grounded in ESM3 Academy articles, with links to the sources. No need to leave this page.

Ask AI about this article ↗