ColabFold is a fast and accessible platform for predicting protein structures and complexes, combining the fast homology search of MMseqs2 with AlphaFold2 or RoseTTAFold. It enables prediction of nearly 1,000 structures per day on a single GPU. ColabFold is open-source and available at https://github.com/sokrypton/ColabFold and its environmental databases at https://colabfold.mmseqs.com. It improves upon AlphaFold2 by using MMseqs2 for faster homology searches and optimized MSA generation, reducing memory usage and enabling faster predictions. ColabFold also includes new environmental databases, such as ColabFoldDB, which expands the diversity of sequences available for structure prediction. It supports both single-chain and complex predictions, and offers a command line interface for batch processing. ColabFold outperforms AlphaFold-Colab and matches AlphaFold2 in accuracy on CASP14 targets. It also matches AlphaFold-multimer on the ClusPro dataset. ColabFold is designed to be accessible to researchers without high-end computing resources, leveraging Google Colaboratory's GPU capabilities. It includes features such as early stopping, optimized model inference, and support for protein complex prediction. ColabFold enables large-scale structure prediction, with the ability to predict the Methanocaldococcus jannaschii proteome in 48 hours with high accuracy. It is available as Jupyter notebooks, a command line tool, and as open-source software. ColabFold improves upon AlphaFold2 by enhancing sequence search, providing tools for modeling complexes, and expanding environmental databases. It is a powerful tool for protein structure prediction, offering high accuracy and speed.ColabFold is a fast and accessible platform for predicting protein structures and complexes, combining the fast homology search of MMseqs2 with AlphaFold2 or RoseTTAFold. It enables prediction of nearly 1,000 structures per day on a single GPU. ColabFold is open-source and available at https://github.com/sokrypton/ColabFold and its environmental databases at https://colabfold.mmseqs.com. It improves upon AlphaFold2 by using MMseqs2 for faster homology searches and optimized MSA generation, reducing memory usage and enabling faster predictions. ColabFold also includes new environmental databases, such as ColabFoldDB, which expands the diversity of sequences available for structure prediction. It supports both single-chain and complex predictions, and offers a command line interface for batch processing. ColabFold outperforms AlphaFold-Colab and matches AlphaFold2 in accuracy on CASP14 targets. It also matches AlphaFold-multimer on the ClusPro dataset. ColabFold is designed to be accessible to researchers without high-end computing resources, leveraging Google Colaboratory's GPU capabilities. It includes features such as early stopping, optimized model inference, and support for protein complex prediction. ColabFold enables large-scale structure prediction, with the ability to predict the Methanocaldococcus jannaschii proteome in 48 hours with high accuracy. It is available as Jupyter notebooks, a command line tool, and as open-source software. ColabFold improves upon AlphaFold2 by enhancing sequence search, providing tools for modeling complexes, and expanding environmental databases. It is a powerful tool for protein structure prediction, offering high accuracy and speed.