Anthropic's latest flagship AI might not have been incredibly costly to train

Anthropic's latest AI model, Claude 3.7 Sonnet, was trained at a cost of "a few tens of millions of dollars" using less than 10^26 FLOPs of computing power, as clarified by Wharton professor Ethan Mollick. This cost is relatively low compared to the training expenses of other top models from 2023, such as OpenAI's GPT-4, which exceeded $100 million, and Google's Gemini Ultra, estimated at nearly $200 million. Despite the current affordability, Anthropic CEO Dario Amodei anticipates future AI models will require billions of dollars for development.

The decreasing cost of training state-of-the-art models highlights the evolving landscape of AI development, though it doesn't account for additional expenses like safety testing and research. As the industry shifts towards "reasoning" models that tackle complex problems over extended periods, the operational costs of running these models are expected to increase. The article, written by TechCrunch senior reporter Kyle Wiggers, underscores the ongoing advancements and financial dynamics within the AI sector.

Key takeaways:

Anthropic's Claude 3.7 Sonnet AI model cost "a few tens of millions of dollars" to train, using less than 10^26 FLOPs of computing power.
Training costs for state-of-the-art AI models are becoming relatively cheaper, with Claude 3.5 Sonnet's predecessor also costing a few tens of millions of dollars.
In contrast, OpenAI's GPT-4 and Google's Gemini Ultra models cost over $100 million and close to $200 million, respectively, to train.
Future AI models are expected to cost billions of dollars, with increasing computing costs due to the industry's shift towards "reasoning" models.

Anthropic's latest flagship AI might not have been incredibly costly to train | TechCrunch

Key takeaways:

Comments (0)

Newsletter