AI model cost per million tokens 2025

The pricing landscape for AI models in 2025 reveals a wide range of costs, with o1 commanding the highest price at

U.S. dollars per million tokens. Claude 3.5 Sonnet follows at U.S. dollars, while other models like Llama 3.1 offer more affordable options at U.S. dollars. This pricing spectrum reflects the evolving market for AI capabilities and the strategic positioning of different providers.

Training costs and environmental impact

The pricing differences among AI models can be partly attributed to their development costs. In 2024, Llama 3.1-405B led with an estimated training cost of

million U.S. dollars, surpassing GPT-4 at million U.S. dollars. These substantial investments in model development are reflected in their market pricing. Additionally, the environmental impact of AI training has grown significantly, with carbon emissions increasing from a mere tons for AlexNet in 2012 to a staggering tons for Llama 3.1-405B in 2021.

Generative AI spending trends

The pricing trends align with broader industry spending patterns on generative AI. In 2024, expenditure on foundational models saw a nearly

increase compared to 2023, outpacing growth in other categories. This surge in investment reflects the expanding possibilities and applications of generative AI in various business sectors. As companies continue to explore and implement AI solutions, the demand for advanced models is likely to influence pricing strategies in the coming years.