Many companies are finding AI costs to be high, questioning the validity of comparing models by $X per 1M tokens. Differences in tokenization methods and how tokens contribute to overall performance significantly impact actual costs, making straightforward comparisons unreliable.
The rise of AI has led to increasing scrutiny of pricing models, with many businesses shocked by their API bills. A common mistake is comparing models based solely on their cost per million tokens, which can be misleading.
Each AI provider has its own tokenizer, which splits text into tokens differently. For instance, the same input may result in varying token counts across different models, such as GPT-4 and GPT-4o. These discrepancies create challenges in fairly comparing costs between models.
Recent modifications to tokenizers, like those made by Anthropic for their Claude model, can drastically affect token counts. A token count increase of over 30% per request due to tokenizer changes can make it appear as if prices have risen significantly, complicating cost assessments.
The true cost of AI usage also depends on how many useful outputs are generated per token, rather than just the price. A token might reflect a lengthy process of 'thinking' that isn't visible in the final output, leading to increased costs.
In comparing a selection of top AI models from both American and Chinese labs, it becomes evident that pricing structures don’t give a complete picture of cost efficiency. For example, GPT-5.5, despite being more expensive per token, may offer lower overall costs per completed task than cheaper alternatives, emphasizing the importance of a comprehensive understanding of AI costs.
✨ This summary was generated by AI from the outlets' reporting listed below. It is not independently verified and may contain errors — check the original sources. How BrevFeed works →
Many companies are finding AI costs to be high, questioning the validity of comparing models by $X per 1M tokens. Differences in tokenization methods and how tokens contribute to overall performance significantly impact actual costs, making straightforward comparisons unreliable.