ESMFold2 Biohub Guide

ESMFold2 is part of Biohub's protein biology world model release. The official Biohub model page should be treated as the first stop for current model positioning, quickstart links, API references, and model-card destinations.

This site uses Biohub sources as factual anchors, but it is not operated by Biohub and does not present itself as an official support channel.

Before publishing API snippets, verify the current package install command, model ID, client imports, folding configuration, and Hugging Face model-card link from the official Biohub or GitHub pages.

The names of client classes and helper functions can change faster than SEO pages. This is why generated code on this site stays conservative and labels code as starter scaffolding.

The Biohub release frames ESMC as the model foundation, ESMFold2 as the all-atom structure prediction model, and ESM Atlas as a large-scale structure and sequence resource.

For SEO, this relationship is useful because users often search for Biohub, ESMFold2, and ESM Atlas together. The page should answer that relationship without turning the site into a broad Biohub directory.

The model page is the best starting point because it gathers the public-facing model description, quickstart entry points, paper link, GitHub link, Hugging Face link, and API guidance in one place.

The GitHub repository is the best implementation reference because it contains package guidance, cookbook examples, and code paths that are more actionable than a news announcement. The announcement is still useful for the release narrative and for understanding the ESMC, ESMFold2, and ESM Atlas bundle.

Nature coverage is useful as third-party reporting, especially for claims about the scale of the atlas, but it should not replace direct Biohub sources for current code or model identifiers.

This site uses Biohub facts to orient users, then turns them into practical guides and browser-local tools. It does not copy official notebooks, pretend to be Biohub support, or present old code snippets as current facts.

When the page mentions open code or model weights, it keeps the license statement together with acceptable-use boundaries. That wording is intentionally more careful than simply saying "free open source," because scientific tooling can carry platform and biosafety constraints.

Researchers who only need the official model should open the Biohub model page first. Developers should continue to the GitHub guide and API guide. SEO visitors comparing names should read the ESMFold2 vs ESMFold page before acting on old ESMFold search results.

Users who want hosted prediction should not treat this Track A site as a prediction service. Hosted prediction is a future Phase 2 layer that must use real ESMFold2 inference, usage-based pricing, data retention rules, and biosafety screening.

When writing about Biohub and ESMFold2, use plain attribution. The site can say Biohub released ESMFold2, Biohub provides the model page, and Biohub links official implementation resources. It should not imply partnership, endorsement, or official status.

For performance claims, the wording should name the reporting source. If Biohub reports a selected comparison result, the page should say Biohub reports it. If Nature reports scale or context, the page should name Nature. This keeps the content useful without overstating certainty.

That approach also helps the page absorb future updates. If Biohub changes a model card, adds a paper link, or revises API examples, the page can be updated cleanly without unwinding broad unsupported claims.

This launch page is written as a third-party guide, not as an official Biohub page. It links to Biohub, GitHub, Colab, model-card, paper, and reporting sources so readers can verify current details before copying code or making technical decisions.

The free Track A site does not collect protein sequences, private Biohub tokens, prediction outputs, or uploaded structure files. Browser-local tools can prepare starter code and examples, but hosted prediction, stored results, queues, and credits remain Phase 2 features.

Predicted structures should not be described as experimental validation, clinical evidence, or drug-development proof. When this page mentions model scope or comparisons, it keeps claims tied to official or clearly attributed sources.

If a source changes after this page is published, the source should win. The page should be updated with a new verification date instead of preserving stale client names, model IDs, pricing assumptions, or unsupported benchmark language.