
The cost of processing a million AI tokens is dropping roughly ten times every year, with DeepSeek V4 Flash at $0.14, OpenAI’s GPT-5.6 Luna at $0.20, and Meta’s Muse Spark 1.2 at…
The cost of processing a million AI tokens is dropping roughly ten times every year, with DeepSeek V4 Flash at $0.14, OpenAI’s GPT-5.6 Luna at $0.20, and Meta’s Muse Spark 1.2 at $1.25. Yet inside enterprises, the focus has shifted from cheap intelligence to value. Tech leaders are now asking which tasks deserve expensive frontier models and which can use cheaper alternatives. They are putting budgets on AI features, tracking token consumption per team, and insisting that every use case justify its cost.

At Eightfold AI, a proposed AI feature had its projected cost cut from $2,000 to $400 before approval. According to Anuraag Kochhar of ShepHertz, 25-40% of Indian enterprises are piloting open-weight models for routine tasks, cutting inference costs by 30-70%, but compliance and governance remain barriers. The new rule: prove the return before spending the token.
The narrative that cheaper AI automatically means more corporate adoption is misleading. While token prices are plummeting, enterprises are tightening budgets, not loosening them. They are asking whether each rupee spent on tokens drives revenue or cuts costs, and rejecting use cases that do not. The real test will come when firms start publishing AI-spend disclosures. Until then, the question is: how many companies are actually tracking where their tokens go? The article mentions fewer than two in three. That number will decide whether the AI boom is efficient or wasteful.
Source: inc42.com
This story was synthesised by AI from the source linked above.