Anthropic’s latest flagship AI model, Claude 3.7 sonnet, can only be trained with less than 10^26 computing power.
According to Walton professor Ethan Mollick, he expressed clarifications he received from Anthropic’s PR in his X post on Monday. “Human contacted me, and he told me that sonnet 3.7 would not be considered a 10^26 flop model and it cost tens of millions of dollars.” He wrote“Although the future model will be bigger.”
TechCrunch extended everyone’s confirmation, but no response was received as of the time of publication.
Assuming that Claude 3.7 sonnet does only cost “ten millions of dollars” to train, rather than considering the associated costs, this suggests that publishing state-of-the-art models has become relatively cheap. Claude’s sonnet predecessor, released in fall 2024 Similarly, training costs tens of millions of dollarsDario Amodei, the CEO of anthropomorphism, revealed in a recent article.
These totals compare to the training price tags for the top 2023 models. Openai spent more than $100 million to develop its GPT-4 model. according to To Openai CEO Sam Altman. Meanwhile, Google spent nearly $200 million to train its Gemini supermodel, a Stanford University study Estimated.
Having said that, Amodei hopes for future AI models Costs billions of dollars. Of course, training costs won’t capture work like security testing and basic research. In addition, with the “inference” model adopted by the artificial intelligence industry, it is used to solve problems Long timethe computational cost of running the model may continue to rise.
