Stanza is an open-source Python natural language processing (NLP) toolkit that supports 66 human languages. It features a fully neural pipeline for text analysis, including tokenization, multi-word token expansion, lemmatization, part-of-speech and morphological feature tagging, dependency parsing, and named entity recognition. The toolkit has been trained on 112 datasets, including the Universal Dependencies treebanks and other multilingual corpora, and shows strong generalization across languages. It also includes a Python interface to the widely used Java Stanford CoreNLP software, extending its functionality to include coreference resolution and relation extraction. Source code, documentation, and pretrained models for 66 languages are available at https://stanfordnlp.github.io/stanza/. Stanza's neural pipeline is designed to process raw text into annotations, including tokenization, multi-word token expansion, lemmatization, part-of-speech and morphological feature tagging, dependency parsing, and named entity recognition. It supports multilingual processing and achieves state-of-the-art performance on various tasks. The toolkit also features a Python client interface to the Java Stanford CoreNLP software, allowing access to additional NLP tools. Stanza is fully open source and provides pre-trained models for all supported languages and datasets. It is designed to be run on different hardware devices, with CUDA devices used when available, or CPUs otherwise. The toolkit supports automated model download via Python code and pipeline customization with processors of choice. Annotation results can be accessed as native Python objects for flexible post-processing. Stanza's CoreNLP client interface allows users to annotate text with the CoreNLP server, which is transparent to the user. The client communicates with the server through RESTful APIs, and annotations are transmitted in Protocol Buffers and converted back to native Python objects. Users can also specify JSON or XML as annotation formats. Stanza also provides an interactive web-based demo for visualizing documents and their annotations. This demo runs the pipeline interactively and visualizes the results with the Brat rapid annotation tool. It is available at http://stanza.run/. Stanza provides command-line interfaces for users to train their own customized models. Users need to prepare training and development data in compatible formats, such as CoNLL-U for the Universal Dependencies pipeline and BIO format column files for the NER model. Stanza has been evaluated on 112 datasets, including the Universal Dependencies v2.5 treebanks and NER datasets. It achieves high performance on these datasets, with Stanza's language-agnostic architecture adapting well to different languages and genres. It also performs well in NER tasks, achieving higher or close F1 scores compared to other tools like FLAIR and spaCy. Stanza is compared with existing toolkits in terms of speed and performance. It is slower than spaCy due to its extensive use of accurate neural models, but it is still competitive when compared against toolkits of similar accuracy,Stanza is an open-source Python natural language processing (NLP) toolkit that supports 66 human languages. It features a fully neural pipeline for text analysis, including tokenization, multi-word token expansion, lemmatization, part-of-speech and morphological feature tagging, dependency parsing, and named entity recognition. The toolkit has been trained on 112 datasets, including the Universal Dependencies treebanks and other multilingual corpora, and shows strong generalization across languages. It also includes a Python interface to the widely used Java Stanford CoreNLP software, extending its functionality to include coreference resolution and relation extraction. Source code, documentation, and pretrained models for 66 languages are available at https://stanfordnlp.github.io/stanza/. Stanza's neural pipeline is designed to process raw text into annotations, including tokenization, multi-word token expansion, lemmatization, part-of-speech and morphological feature tagging, dependency parsing, and named entity recognition. It supports multilingual processing and achieves state-of-the-art performance on various tasks. The toolkit also features a Python client interface to the Java Stanford CoreNLP software, allowing access to additional NLP tools. Stanza is fully open source and provides pre-trained models for all supported languages and datasets. It is designed to be run on different hardware devices, with CUDA devices used when available, or CPUs otherwise. The toolkit supports automated model download via Python code and pipeline customization with processors of choice. Annotation results can be accessed as native Python objects for flexible post-processing. Stanza's CoreNLP client interface allows users to annotate text with the CoreNLP server, which is transparent to the user. The client communicates with the server through RESTful APIs, and annotations are transmitted in Protocol Buffers and converted back to native Python objects. Users can also specify JSON or XML as annotation formats. Stanza also provides an interactive web-based demo for visualizing documents and their annotations. This demo runs the pipeline interactively and visualizes the results with the Brat rapid annotation tool. It is available at http://stanza.run/. Stanza provides command-line interfaces for users to train their own customized models. Users need to prepare training and development data in compatible formats, such as CoNLL-U for the Universal Dependencies pipeline and BIO format column files for the NER model. Stanza has been evaluated on 112 datasets, including the Universal Dependencies v2.5 treebanks and NER datasets. It achieves high performance on these datasets, with Stanza's language-agnostic architecture adapting well to different languages and genres. It also performs well in NER tasks, achieving higher or close F1 scores compared to other tools like FLAIR and spaCy. Stanza is compared with existing toolkits in terms of speed and performance. It is slower than spaCy due to its extensive use of accurate neural models, but it is still competitive when compared against toolkits of similar accuracy,