Cohere, a company specializing in artificial intelligence technologies, has announced the release of its new model, Embed 4, designed to enhance enterprise search by analyzing both textual and visual data.
This model is capable of processing documents up to 200 pages long and can understand and analyze an extensive range of informationāup to 128,000 tokens in a single instance.
Thus, it becomes particularly valuable for organizations working with large volumes of unstructured data, such as text files, emails, images, and videos.
Embed 4 was developed specifically for industries that require high levels of confidentiality and regulatory compliance, including healthcare and finance. It offers high accuracy in extracting information from complex documents such as medical reports and legal contracts.
Unlike previous models, Embed 4 eliminates the need for extensive data preprocessingāsuch as reformatting documentsābefore analysis. According to the company, it can still operate effectively even when documents contain typos or irregular formatting.
Customer testing, including that of the e-commerce platform Agora, has shown a 47% improvement in search result accuracy compared to earlier versions.
Agoraās team attributes this improvement to Embed 4ās ability to unify both textual and visual product descriptions into a single numerical format that the system can understand, resulting in faster searches and lower operational costs.
Another key feature of the model is its ability to compress extracted data, helping reduce storage costs by up to 83% while maintaining output quality.
This is especially beneficial for large enterprises that rely on AI systems to handle tasks such as analyzing financial reports or supporting investment decisions.
Additionally, Embed 4 supports over 100 languagesāincluding Arabic, Japanese, and Koreanāenabling global organizations to search across multilingual documents without difficulty.
The model also integrates with platforms like Microsoft Azure and Amazon SageMaker. It offers deployment options that meet high-security requirements, whether through on-premises storage or private cloud servers.
Furthermore, this development supports a key AI technique known as Retrieval-Augmented Generation (RAG).
RAG helps advanced AI systemsālike modern chatbotsāretrieve the most relevant and up-to-date information from within an organization's data to provide better, more reliable answers, rather than relying solely on pre-trained knowledge.
As automation becomes more integrated into daily business workflows, models like Embed 4 offer a practical way to turn complex, disorganized data into useful, easily searchable information that teams can actually rely on.
