By Krystal Hu
Canadian AI startup Cohere, last valued at $5.5 billion, will focus on building tailored models for enterprise users over larger foundation models, the company told Reuters.
The evolvement of its strategy, laid out in a company letter to its investors on Thursday, comes as many companies are still trying to figure out how to incorporate large language models into their daily work two years since ChatGPT burst onto the scene.
“What we’re hearing from customers is that they don’t just need bigger models to be good at everything. They need models that are actually built for their specific use cases,” Nick Frosst, co-founder of Cohere, said in an interview with Reuters.
Cohere, seen as a competitor to AI labs including OpenAI and Anthropic, says it will continue to develop foundation models, but will focus on other training techniques to improve models, instead of increasing model sizes. While selling Application Programming Interface (NASDAQ:TILE) (API) to its models will remain a small part of Cohere’s offering, the focus is on customized model deployment.
The race to build bigger and better models has fueled an investment boom from startups to big tech. OpenAI, Anthropic and xAI have raised billions to fund the capital-intensive development of frontier AI models.
Headquartered in Toronto and San Francisco, Cohere has raised over $900 million from investors including Nvidia (NASDAQ:NVDA), Cisco (NASDAQ:CSCO), and Innovia Capital.
Cohere has pitched itself as an enterprise focus AI company independent of cloud providers. It has been working directly with customers such as Oracle (NYSE:ORCL) and Fujisu to tailor models for specific needs.
Cohere’s new focus also comes as the industry that had seen breakthroughs by scaling computational power and model size, is seeing diminishing returns from bigger models. AI labs are facing delays in training the new generation of large language models. Ilya Sutskever, co-founder of AI labs Safe Superintelligence (SSI) and OpenAI, told Reuters recently that results from scaling up pre-training have plateaued.
Frosst said that simply increasing model size doesn’t always yield better results. The focus on customization could allow Cohere to be more capital-efficient, reducing the need for computational power. The company is not pursuing artificial general intelligence (AGI) like OpenAI.
“We’re going to work with an enterprise to figure out how we can make the model perfect at their use case, tailor it to that specific needs and get to production, not bank on the AGI future is coming next year,” Frosst said.
