The most recent flagship AI of anthropy may not be incredibly expensive

The latest flagship AI model of anthropic, Clod 3.7 Sonnet, 10 ^ 26 “A few to ten million dollars” to train less than 10 ^ 26 “.

According to Wharton Professor Ethan Mollick, the Anthropik, who was in a writing on Monday, brought a clarity from the PR. “Let me say that Sonnet 3.7 did not consider 10 ^ 26 flop models and communicated with an anthropic that says that it does not cost several Ten Millions” wrote“Though future models are greater.”

TechCrunch reached anthropicle for approval, but the publication did not receive the answer.

Claude 3.7 Sonnet is really a sign of “millions of dollars in a few million dollars”, not to factoring on related expenses, but the most modern models. Claude 3.5 Sonnet’s predecessor, autumn released in 2024, Similarly for the train is several tens of millions of dollarsAnthropic CEO Dario Amodei appeared in the last essay.

In total, the best models of the best models of 2023 are very positive comparable for the labels. To develop the GPT-4 model, Openai spent more than $ 100 million, according to Openai CEO Sam Altman. Meanwhile, Google, Gemini spent about $ 200 million to bring up the Ultra model, Stanford research estimated.

It is said that Amodei is waiting for future AI models billions of dollars costs. Undoubtedly, training costs do not work as security tests and fundamental studies. Moreover, the AI ​​industry embraces the “reasoning” models working on problems Time periodComputing costs of running models will continue to increase.