Introducing Cohere Summarize Beta: A New Endpoint for Text Summarization

Introducing Cohere Summarize Beta: A New Endpoint for Text Summarization


“Artificial intelligence firm Cohere has launched a text summarization endpoint. Built on the firm's foundational language model, the Cohere Summarize beta service allows the condensing of essential information from large documents. The system can process long documents of up to 50,000 characters and includes a series of customizable settings.”

“This tl;dr was generated by the Cohere Summarize product. Don’t believe it, see it in action.

In a world where information overload is the norm, the ability to quickly and accurately distill essential information from documents has become increasingly important for businesses and individuals alike. That's why we are excited to provide early access to the Cohere Summarize – a new endpoint powered by our foundational language model, customized for text summarization.

Summarize beta empowers users to efficiently distill crucial information from lengthy documents and articles. This functionality greatly increases productivity and saves valuable time, making it essential for anyone looking to extract the most essential information quickly and accurately. Gone are the days when users had to worry about crafting the perfect prompt or teaching a model how to generate a summary. With our endpoint, summarizing any article or text is as simple as inputting it into the system, which will automatically generate a summary for you.

Summarizing documents with modern large language models is a powerful technique for language understanding and production. Through our extensive research, which draws on years of experience in the field of Natural Language Processing, coupled with valuable feedback from our early customers, we have identified three key advantages that individuals and companies can benefit from by utilizing our Summarize endpoint.

  • Specialized summarization endpoint: Instead of relying on one of our Large Language Models that excel at both information extraction and creative writing, which may introduce new facts, we've developed a specialized endpoint. The model powering this endpoint is customized specifically for summarization. Gone are the days when users had to worry about crafting the perfect prompt or teaching a model how to generate a summary. While this improves the factuality significantly, we still recommend that users verify the output for any potential factual inconsistencies. Modern large language models are based on sophisticated pattern matching and, despite significant progress, occasionally return wrong facts with high certainty. This can lead to what is known as "hallucinations," where the model generates inaccurate information. As we continue through this beta period, we will be actively working to enhance our factuality.  
  • Summarize longer documents: Summarize lengthy documents up to 50,000 characters, which is equivalent to 18-20 pages in a single-spaced format. This goes beyond the context length of most large language models, making it easy to get high-quality summarization results. Summarizing long documents is a challenge for most models, but with Cohere Summarize, we extract key sentences from the document up to the context length, and then generate a synthetic document composed of the combination of those key sentences. This approach ensures that the model has a comprehensive overview of the entire document from the outset. This feature saves time and resources, especially for businesses that regularly deal with lengthy documents. For example, you can easily summarize "Attention Is All You Need", the seminal 2017 research paper in the field of natural language processing, co-authored by one of our founders, Aidan Gomez.
  • Personalized summaries: We offer customizable settings that allow users to control certain aspects of their summaries, such as the length of the summary and output formats, including paragraph and bullet points. Due to the subjective nature of individual preferences and requirements, the process of summarization can be ambiguous and what qualifies as a good summary may vary. We all have different interests, and read and understand language in different ways. This creates a challenge for generating "personalized summaries" that accurately reflect the needs and interests of different individuals. We are continuously working on making the Summarize endpoint even more personalized by allowing users to specify additional commands to the model expressed in natural language.

Get Started with Cohere Summarize API

Accessing the Cohere Summarize API is easy and user-friendly. You can start exploring the API right away by accessing our Summarize playground, where you can input your own text and receive an instant summary generated by our language model. Alternatively, if you're a developer looking to integrate the Cohere Summarize API into your own application or workflow, you can access our documentation, which provides detailed instructions and code samples. You can also visit our Summarize product page. With our straightforward and accessible tools, getting started with Cohere Summarize has never been easier.

Help us make Cohere Summarize even better

While our beta endpoint may not have all the answers just yet, it's a powerhouse of potential waiting to be unleashed. We are constantly working to enhance it, expanding its capabilities and improving accuracy every day. Cohere Summarize has the potential to become an essential product for forward-thinking innovators, builders, and creators worldwide. We value your thoughts and feedback, so please don't hesitate to share them with us through our Discord Community or by clicking on the chat bubble icon located in the bottom left-hand corner of the playground when you log in. Alternatively, you can reach us at