OpenAI outperforms Chinese models in cost per task (nikvdp.com)

OpenAI's latest model, GPT-6.1 Sol, has emerged as a more efficient choice compared to several Chinese models, particularly DeepSeek V4.1 Flash, when evaluating cost per task rather than price per token. While DeepSeek appears cheaper at first glance—charged per token at seven to eight times less than Sol—its performance requires substantially more tokens to complete similar tasks, ultimately resulting in higher costs for users. This disparity highlights the importance of understanding how different models define and utilize tokens, as the actual cost for completing tasks can vary greatly based on model efficiency. This development is significant for the AI/ML community as it illustrates the critical concept of "cost per task," which better reflects the real-world implications of using different models. As data from Artificial Analysis shows, OpenAI has consistently improved its standing, with its models now dominating the task efficiency frontier. Despite the clear advantages of OpenAI's offerings, surprising user preferences for Chinese models like DeepSeek suggest that brand perception and marketing may significantly influence model selection, potentially leading to less rational usage patterns in the marketplace. As the landscape evolves, the emphasis on practical cost-effectiveness will likely drive further advancements in model efficiency and performance.