The MTEB Arena, the first-ever platform dedicated to evaluating embedding models, has been launched. This innovative initiative allows users to dynamically test and vote on the quality of various embedding models, which have shown surprising variance in MTEB scores. The arena features 15 models from prominent organizations, including OpenAI, Google, Cohere, Voyage AI, Jina AI, SF Research, and Nomic AI, competing in three tasks: retrieval, clustering, and semantic textual similarity (STS). The project is a collaborative effort led by Muennighoff, with significant contributions from Vaibhav Adlakha and Siva Reddy, and is supported by sponsorships from organizations such as ServiceNow Research. This new platform aims to enhance the evaluation process in the field of embeddings and promote open science.