"We will provide enterprise AI solutions that do not replace systems customers have already invested in, but complement them to increase effectiveness and make them cheaper to use."

Frank O'Dowd (프랭크 오다우드), chief revenue officer at enterprise AI company Cohere, stressed that point on Tuesday as he outlined what he called Cohere's differentiators in the enterprise AI market. He also announced plans to set up local units in South Korea and Japan and to double the company's technology and sales headcount in the Asia-Pacific region by the end of this year.

At the briefing, O'Dowd highlighted that Cohere has pursued an enterprise AI strategy focused on sovereign AI from the outset, and that other companies have only recently begun to follow its approach.

"Cohere provides services across all major cloud providers," he said. "At the same time, companies in regulated industries such as finance, manufacturing, telecommunications, healthcare and pharmaceuticals can also use it on an on-premise basis in their own data centres." He added, "Unlike competitors, we do not build platforms to increase model sales." He stressed that the company is building a platform that can use models in any environment, including on-premise, data centres and the cloud.

According to the company, Cohere's enterprise AI platform North provides not only models but also embeddings, rerank (Cohere Rerank) and translation models.

North supports building inference and agentic AI solutions using both frontier models and open-source models, regardless of whether the environment is cloud-based or on-premise. Customers can select and manage the most efficient model by task and cost through an agent catalogue.

Cohere Compass, a search solution, supports lowering token costs by using rerank functions at the stage before frontier-model search, based on embeddings, rerank and translation models. Cohere Rerank is a rerank-only model provided by Cohere that takes a list of documents initially found by a search system, reassesses how relevant they are to the actual question and reorders them.

Cohere also plans to release a new large model, Command A Plus, with 1 trillion parameters by the end of the year.

Value for money was another keyword O'Dowd emphasised. "Cohere models run with less hardware than other companies' models, so computational efficiency is high," he said. "Thanks to this efficiency, agentic AI can be implemented in a range of environments, including regions where access to GPUs is limited." On translation, he also expressed confidence, saying, "In Arabic benchmarks, it delivered performance on par with three top frontier models."