How much does it cost to build a LLM such as ChatGPT?
According to recent estimates, the cost to build and train large language models (LLMs) like ChatGPT has increased significantly:
-
ChatGPT-4, the latest version, had an estimated technical creation cost of $41 million to $78 million[5]. OpenAI CEO Sam Altman has stated it cost over $100 million when including all expenses.
-
Earlier versions were much less expensive. ChatGPT-3 cost only around $2 million to $4 million to train in 2020[5].
-
Google's Gemini LLM reportedly cost between $30 million and $191 million to train, not including staff salaries[5].
-
When factoring in ongoing operational costs, ChatGPT's daily expenses are estimated to be $100,000 to $700,000, or $3 to $21 million per month[2].
-
These high costs are primarily due to the massive computing power required, use of specialized hardware like GPU clusters, and the expertise of highly paid AI researchers and engineers.
The extreme costs of developing cutting-edge LLMs have made it challenging for academic or public institutions to keep pace with private tech companies in AI research and development[5].
Citations:
[1] https://www.cerait.com/cost-embedding-chatgpt-business-website.html
[2] https://talkai.info/blog/inside_chatgpt_economy/
[3] https://usmsystems.com/how-much-does-it-cost-to-develop-a-chatbot-like-chatgpt/
[4] https://yellow.systems/blog/chatgpt-application-development-cost
[5] https://www.forbes.com/sites/katharinabuchholz/2024/08/23/the-extreme-cost-of-training-ai-models/
[6] https://www.ptolemay.com/post/how-much-does-chatgpt-integration-cost-and-pay-off
[7] https://www.yahoo.com/tech/chatgpt-cost-bomb-openais-losses-125101043.html
[8] https://www.reddit.com/r/ChatGPT/comments/1c5ri5o/if_chatgpt_costs_700000_to_run_per_day_how_do/