Feb. 26 (Bloomberg) -- According to Wharton professor Ethan Mollick on social media platform X on Monday, artificial intelligence company Anthropic The latest flagship model of the Claude 3.7 Sonnet's training cost only "tens of millions of dollars".
Mollick said he learned from Anthropic's public relations department that Sonnet 3.7 is not a 10^26 FLOP model and that it cost "tens of millions of dollars" to train, though future models will be much larger.
As of this writing, Anthropic has not yet responded to this.
If the training costs for Claude 3.7 Sonnet are indeed in the "tens of millions of dollars", regardless of other associated costs, this suggests that the cost of releasing state-of-the-art models is becoming relatively inexpensive, as mentioned in a recent paper by Anthropic CEO Dario Amodei. paper, Dario Amodei, CEO of Anthropic, mentioned that Claude 3.5 Sonnet, due for release in the fall of 2024, will also cost tens of millions of dollars to train.
Compared to the cost of training the top models in 2023, this figure seems quite competitive. According to OpenAI CEO Sam Altman, OpenAI invested more than $100 million to develop the GPT-4 model. Meanwhile, a Stanford University study estimated that Google spent close to $200 million training its Gemini Ultra model.
