SWISS-MODEL is a fully automated server for homology modelling of protein structures and complexes. It has been continuously developed for 25 years and recently extended to model homo- and heteromeric complexes. The server uses a new modelling engine, ProMod3, and a new local model quality estimation method, QMEANDisCo. SWISS-MODEL generates approximately 3000 models per day and is one of the most widely used structure modelling servers. It is continuously evaluated and compared with other state-of-the-art servers. The server participates in the CAMEO project, which provides a continuous automated assessment of predictions from various modelling servers. SWISS-MODEL also provides cross-links to other resources and databases for exploring information on target proteins. The server allows for customization of steps in the modelling workflow and provides a detailed documentation. The SWISS-MODEL Template Library (SMTL) is a curated template library updated weekly based on new PDB releases. The SMTL contains coordinates for 92,474 unique protein sequences. The SWISS-MODEL Repository contains 1,067,355 models from SWISS-MODEL and 129,416 structures from PDB with mapping to UniProtKB. The server also provides a specialised server for antibody modelling. The server uses Python and Django with JavaScript and jQuery for the front-end, and Python, C++ and OpenStructure for the back-end functions. The server provides two interactive JavaScript/WebGL molecule viewers for protein structure visualization. The server has been updated with a new modelling engine, ProMod3, which has been shown to significantly improve model accuracy compared to the previous engine. The server has also been improved in terms of model quality estimation. The server has been tested in the CAMEO experiment and has consistently ranked among the top-modelling servers for several crucial modelling aspects. The server has been shown to excel at model quality for binding sites, high-quality models, and quaternary structure prediction. The server has also been used to model the Ferredoxin/Ferredoxin-NADP(+) Reductase complex, demonstrating its ability to model complex protein structures. The server has been shown to provide accurate local confidence estimates of the returned models. The server has been shown to be effective in handling ambiguous or conflicting information present in available structural data, which is crucial for the development of stable and fully automated pipelines. The server has also been shown to provide additional criteria to improve the automatic identification of templates, leading to more accurate models and a biologically meaningful representation of their oligomeric state. The server has also been shown to provide more accurate models and realistic error estimates.SWISS-MODEL is a fully automated server for homology modelling of protein structures and complexes. It has been continuously developed for 25 years and recently extended to model homo- and heteromeric complexes. The server uses a new modelling engine, ProMod3, and a new local model quality estimation method, QMEANDisCo. SWISS-MODEL generates approximately 3000 models per day and is one of the most widely used structure modelling servers. It is continuously evaluated and compared with other state-of-the-art servers. The server participates in the CAMEO project, which provides a continuous automated assessment of predictions from various modelling servers. SWISS-MODEL also provides cross-links to other resources and databases for exploring information on target proteins. The server allows for customization of steps in the modelling workflow and provides a detailed documentation. The SWISS-MODEL Template Library (SMTL) is a curated template library updated weekly based on new PDB releases. The SMTL contains coordinates for 92,474 unique protein sequences. The SWISS-MODEL Repository contains 1,067,355 models from SWISS-MODEL and 129,416 structures from PDB with mapping to UniProtKB. The server also provides a specialised server for antibody modelling. The server uses Python and Django with JavaScript and jQuery for the front-end, and Python, C++ and OpenStructure for the back-end functions. The server provides two interactive JavaScript/WebGL molecule viewers for protein structure visualization. The server has been updated with a new modelling engine, ProMod3, which has been shown to significantly improve model accuracy compared to the previous engine. The server has also been improved in terms of model quality estimation. The server has been tested in the CAMEO experiment and has consistently ranked among the top-modelling servers for several crucial modelling aspects. The server has been shown to excel at model quality for binding sites, high-quality models, and quaternary structure prediction. The server has also been used to model the Ferredoxin/Ferredoxin-NADP(+) Reductase complex, demonstrating its ability to model complex protein structures. The server has been shown to provide accurate local confidence estimates of the returned models. The server has been shown to be effective in handling ambiguous or conflicting information present in available structural data, which is crucial for the development of stable and fully automated pipelines. The server has also been shown to provide additional criteria to improve the automatic identification of templates, leading to more accurate models and a biologically meaningful representation of their oligomeric state. The server has also been shown to provide more accurate models and realistic error estimates.