Cohere

Cohere is a Canadian company providing large language models (LLMs) for developers and enterprise. Cohere builds LLMs pre-trained on billions of words that work on public, private, or hybrid clouds. Cohere's LLMs can be accessed via API or the "Playground," a visual interface that does not require coding experience. Users can customize models to their needs and train them on their data, simplifying the deployment of use-case specificuse-case-specific LLMs. Cohere handles the neural network architecture and the large amount of initial training data required. Cohere offers access to both generation models and representation models, models that return an embedding vector for the input text.

Use cases with Cohere's LLMs include the following:

• Semantic search—text search across documents in over 100one hundred languages
• Generate—producingGenerate—text textproduction for a range of documents (product descriptions, blog posts, articles, marketing copy, etc.) utilizing scalable generative AI tools
• Embed—managed embedding model in over 100one hundred languages
• Rerank—a semantic boost improving the search quality of any keyword or vector search system.

Cohere was founded in 2019 by Aidan Gomez (CEO), Ivan Zhang, and Nick Frosst. Gomez and Frosst were previously part of the Toronto-based Google Brain Team. Gomez was a co-author of the Google paper introducing the transformer neural network architecture. Zhang was previously a founding engineer at Cortex and worked with Gomez as part of the FOR.ai research team. The company emerged from stealth in May 2021, announcing it had partnered with three North American companies, including Toronto AI chatbot startupstart-up Ada. In November 2021, Cohere announced a multiyear partnership with Google Cloud. In May 2023, Cohere announced LLM University, a set of learning resources for natural language processing. Headquartered in Toronto, the company has expanded with offices in Palo Alto, San Francisco, and London.

Cohere offers access to a range of models of different sizes. Large models are more capable of complex tasks, while smaller models have faster response times and are less expensive to run.

Cohere's generation models include:

• Command—the company's most capable generative model, well suited for challenging tasks including complex extraction, rewriting, question-answering, summarization, conversation, and brainstorming.

• Command-light—a tradeoff between power and speed, command-light is capable of performing tasks like generating marketing ad-copy, extracting key entities from text, or powering conversational agents.

• Embed-english-v2.0—Cohere's largest and most capable representation model ideal for few shot classification tasks in both single labelsingle-label and multi-label scenarios. Large embeddings have 4096 dimensions.

On September 7, 2021, Cohere announced $40 million in Series A funding led by Index Ventures, with participation from Section 32, and Radical Ventures. Several well-known researchers in AI, such as Geoffrey Hinton, Fei-Fei Li, Pieter Abbeel, and Raquel Urtasun, also took part in the funding round. Co-foundersCofounders Aidan Gomez (CEO) and Nick Frosst had previously worked with Hinton at Google Brain. Mike Volpi, a partner at Index Ventures, joined the board of Cohere.

In February 2023, Reuters reported Cohere was in talks to raise hundreds of millions of dollars in a round that would value the company at over $6 billion. On March 7, 2023, Cohere was announced as one of the first four startupsstart-ups to receive funding from Salesforce Ventures' new $250 million generative AI investment fund. While the figure received is unknown, Cohere will bring its technology to Salesforce’s enterprise customers as part of the deal.