AlphaFold Server runs Google DeepMind's AlphaFold 3 model in the browser for free, predicting protein, DNA, RNA and ligand complex structures with no install. Pick it for a non-commercial research question about a structure. It is capped at 30 jobs a day per account and its terms forbid any commercial use or using outputs to train a competing model.
Other good fits
Modelling a nucleic acid structure or a complex involving a common ligand, metal ion or post-translational modification.
Academic, non-profit or educational research where a fast, no-install structure prediction is useful groundwork.
+ Pros
Read pros
AlphaFold Server gives anyone with a Google account a free browser interface to AlphaFold 3, Google DeepMind and Isomorphic Labs' model for predicting the 3D structure of biomolecules. It goes beyond single-protein prediction: it can model proteins, DNA, RNA, common ligands (from ATP to chlorophyll), metal ions, and chemical modifications like glycosylation, all in a single complex, with no software installation and no need for your own compute.
Pick it when
You want a fast structural hypothesis for a protein or complex before investing in experimental structure determination.
Your question involves a protein interacting with a ligand, nucleic acid or modification that a single-protein-only predictor would miss.
You're doing non-commercial academic, government or educational research and can work within the daily job quota.
− Cons
Read cons
AlphaFold Server is restricted to non-commercial use by individuals, universities, non-profits, research institutes, educational and government bodies (with journalism as the one stated exception); a commercial company, or research conducted on behalf of one, is out of scope under its terms. Each user is limited to 30 jobs per day, refreshing daily, with no way to buy more capacity through the server itself. Outputs cannot be used to train machine learning models for biomolecular structure prediction, and every prediction is still a hypothesis: a low-confidence region is a signal to verify experimentally, not a fact.
Pick something else when
The work is commercial, or done on behalf of a commercial organisation: AlphaFold Server's terms exclude that use case; a commercial licence or a different provider is needed instead.
You need more than 30 predictions a day or want to run structure prediction as an automated pipeline: that requires a different deployment of the underlying model, not the free web server.
You need general research literature to support the biological question behind a structure: pair this with PubMed, not a substitute for it.
Avoid if
You need more than 30 predictions in a day; the daily quota resets but does not carry over or scale up.
You plan to use the structures to train or build a competing structure-prediction model; that is explicitly prohibited.
Workflow
Sign in with a Google account → define the components of your prediction (protein sequence, nucleic acid, ligand, ions or modifications) → submit the job, which counts against your daily quota → review the predicted structure and its confidence scores once it completes → download or save the result, keeping in mind the output terms of use before publishing or sharing it.
Cost
AlphaFold Server is completely free to use, with no paid tier. There is nothing to install or maintain; the only limits are the daily job quota and the scope of what you're allowed to use it for. Prices checked 23 September 2026: no fee is charged for any tier of access; the constraint is the usage quota, not money. Signing in with a Google account is the only setup step, and the running effort is tracking your daily job count and reading the confidence output carefully rather than treating every prediction as a finished answer.
Restrictions
Non-commercial use only, by individuals and non-commercial organisations (universities, non-profits, research institutes, educational and government bodies), with journalism as the sole listed exception. The service caps each user at 30 jobs per day, and its terms prohibit using outputs to train or build a competing biomolecular structure-prediction model.
First project
Documentation-based suggestion — not run here.
Read first project
Predict the structure of a protein-ligand complex and read the confidence scores before trusting the model. The book A first structure prediction walks through this.
Pick a protein sequence and a small-molecule ligand relevant to your question.
Submit them together as one AlphaFold Server job.
Once it completes, look at the per-residue and per-interface confidence scores, not just the picture of the structure.
Compare the predicted binding pose against anything already known about the interaction from the literature.
Note which parts of the structure have low confidence and treat those as open questions, not answers.
Done when you can say specifically which parts of the prediction are well-supported by the confidence scores and which are not, rather than accepting the whole structure at face value.
Links & evidence
Documentation-based · not hands-on · Sep 23, 2026
alphafoldserver.com itself renders via JavaScript and could not be read with a text fetch, so the current daily quota, non-commercial restriction and supported molecule types rest on EMBL-EBI's official 2026 AlphaFold Server training material, cross-checked against news coverage of the quota's history (10, then 20, now 30 jobs/day) from Science and GEN. Nothing was submitted to AlphaFold Server for this review; the first project is a suggested exercise.
EMBL-EBI: AlphaFold Server, your gateway to AlphaFold 3 · Sep 23, 2026 — States the current 30 jobs/day limit refreshing at midnight, confirms free access via Google sign-in, non-commercial use restriction, and lists supported molecule types (proteins, DNA, RNA, ligands, metal ions, modifications) with a 5,000-token limit per prediction.
GEN: AlphaFold 3 Angst - Limited Accessibility Stirs Outcry · Sep 23, 2026 — Documents the quota's history: a 10 requests/day limit at launch, later raised to 20, and the stated non-commercial restriction on the public web server.
AlphaFold Server Prohibited Use Policy · Sep 23, 2026 — Official policy page (URL confirmed reachable); prohibits commercial use and using outputs to train competing structure-prediction models, corroborated by EBI's training material and secondary reporting.
Further review notes & alternatives
Alternatives and trade-offs
PubMed is the natural companion for checking what is already known about a structure or interaction before or after a prediction, not a substitute for one.
Biopython is useful for scripting around AlphaFold Server's output, for example parsing the predicted structure file (Bio.PDB) as part of a larger pipeline.
A locally run or commercially licensed structure-prediction pipeline is the option when the work is commercial or exceeds the free daily quota; that is a separate setup from this free web server.
AlphaFold Server is a free web interface to Google DeepMind's AlphaFold 3 model for predicting biomolecular structures, covering proteins, DNA, RNA, common ligands, metal ions and chemical modifications. It is restricted to non-commercial use by individuals and academic or non-profit organisations, capped at 30 jobs per user per day, and outputs may not be used to train competing structure-prediction models.
Google DeepMind and Isomorphic Labs · free · cloud
AlphaFold 3 structure prediction
protein/DNA/RNA/ligand complex modelling
metal ion and post-translational modification modelling