New Study Reveals True Costs of AI Models: Price Isn’t Everything
A recent analysis shows that businesses looking for affordable AI solutions may need to rethink their approach. For some time, Chinese AI models have been viewed as budget-friendly alternatives to popular options from OpenAI and Anthropic. However, new research suggests that choosing the cheapest option may lead to higher overall costs.
The study conducted by AlphaSense, an AI-driven market intelligence platform, found that OpenAI’s GPT-5.6 Sol and Anthropic’s Opus 4.8 outperformed Chinese models like Kimi K3 and GLM-5.2 in complex financial analysis tasks, both in quality and cost.
The Cost of Quality
At first glance, Chinese AI models seem to offer a better deal, with rates as low as $15 per million tokens for Kimi K3. In comparison, Anthropic’s Opus 4.8 and OpenAI’s GPT-5.6 Sol cost $25 and $30 respectively. But, as the study revealed, token pricing only tells part of the story.
AlphaSense tested 246 financial analysis tasks, such as reviewing earnings calls and SEC filings. They found that while Kimi K3 had lower token costs, OpenAI’s model provided approximately 20% higher quality responses at around 13% lower overall costs. Anthropic’s Opus 4.8 did even better, generating responses 13% richer in quality for about half the cost of Kimi K3.
The key takeaway is that more advanced models tend to use fewer tokens and steps to achieve the same results. This reduces the total cost, even if their token prices appear higher upfront.
Rethinking AI Expenses
As companies look to adopt advanced AI models from leaders like OpenAI and Anthropic, the report suggests they need to consider not just the price per token but also the overall expenses of completing specific tasks, particularly in detailed fields like financial research. More sophisticated AI can provide accurate results more efficiently, justifying their higher initial cost.
The report does not entirely dismiss the potential of open models. Companies that manage their AI needs within their infrastructure can bypass per-token fees, and for simpler tasks—like summarizing emails—more advanced models may not be necessary.
Instead of relying solely on one type of model, the report suggests that companies may save money by using a mix of different models. For instance, AlphaSense’s platform uses a system that directs parts of a question to the most suitable models. This could mean using a more capable model to outline an answer while a less expensive model executes simpler tasks.
Looking Ahead
As businesses evaluate which AI models to use, they are encouraged to move past merely comparing token costs. The competition in the AI industry may shift from who charges the least to who can offer the best value based on completing tasks most efficiently.
