\n<\/p>\n
Anthropic\u2019s newest flagship AI model, Claude 3.7 Sonnet, cost \u201ca few tens of millions of dollars\u201d to train using less than 10^26 FLOPs of computing power. <\/p>\n
That\u2019s according to Wharton professor\u00a0Ethan Mollick, who in an X post on Monday relayed a clarification he\u2019d received from Anthropic\u2019s PR. \u201cI was contacted by Anthropic who told me that Sonnet 3.7 would not be considered a 10^26 FLOP model and cost a few tens of millions of dollars,\u201d he wrote<\/a>, \u201cthough future models will be much bigger.\u201d<\/p>\n TechCrunch reached out to Anthropic for confirmation but hadn\u2019t received a response as of publication time.<\/p>\n Assuming Claude 3.7 Sonnet indeed cost just \u201ca few tens of millions of dollars\u201d to train, not factoring in related expenses, it\u2019s a sign of how relatively cheap it\u2019s becoming to release state-of-the-art models. Claude 3.5, Sonnet\u2019s predecessor, released in fall 2024, similarly cost a few tens of millions of dollars to train<\/a>, Anthropic CEO Dario Amodei revealed in a recent essay.<\/p>\n Those totals compare pretty favorably to the training price tags of 2023\u2019s top models. To develop its GPT-4 model, OpenAI spent more than $100 million, according<\/a> to OpenAI CEO Sam Altman. Meanwhile, Google spent close to $200 million to train its Gemini Ultra model, a Stanford study estimated<\/a>.<\/p>\n That being said, Amodei expects future AI models to cost billions of dollars<\/a>. Certainly, training costs don\u2019t capture work like safety testing and fundamental research. Moreover, as the AI industry embraces \u201creasoning\u201d models that work on problems for extended periods of time<\/a>, the computing costs of running models will likely continue to rise.<\/p>\n<\/div>\n