Toronto-based AI company Cohere released a new low-cost AI model that it says requires a fraction of the computing power as models from industry leaders like OpenAI and Anthropic.
What happened: Like a middle schooler learning to use cologne, AI companies are slowly learning that less can be more. While Anthropic’s CEO said some models cost US$100 million just to train, Cohere claims it built its new model for less than $30 million.
- By narrowing its scope and catering to business needs, Cohere was able to build a model that excels at tasks like coding, research, and customer service.
Why it matters: Investors and big technology companies have been operating under the assumption that industry-leading AI models require billions of dollars’ worth of computing power, but companies like Cohere and DeepSeek are starting to challenge that logic.
- China’s DeepSeek sparked a tech stock selloff earlier this year after releasing a model it claimed cost only $6 million and could outperform the industry’s best.
- Cohere’s co-founder Nick Frosst told BetaKit that DeepSeek’s breakthrough proved that the key to AI development is “innovation and efficiency, not excessive compute.”
Zoom out: More efficient AI models could have a ripple effect on some tech stocks. The perceived cost of computing power has helped drive an AI stock rally that has been called “one of the greatest spending binges in human history.”—LA