The AI market is undergoing a fascinating transformation, with a stark divide between commodity inference and the rising costs of frontier models. This shift has significant implications for businesses and users, as the price of AI tokens fluctuates widely, leaving many to question if they're getting value for money. Aman Panjwani, an AI engineer, highlights a remarkable decline in the cost of GPT-4-class model output, from $20 per million tokens in late 2022 to $0.40 today, a 55x reduction in less than four years. This drop has been so dramatic that DeepSeek's R1 reasoning model, launched in January 2025, offered a 97% discount compared to OpenAI's O1 preview just four months earlier. However, this trend has been offset by a surge in the prices of cutting-edge frontier models, with OpenAI doubling the price of GPT-5.5 and Google's Gemini Flash 3.5 arriving at a significantly higher cost than its predecessor.
This dynamic has led to a market split, with commodity inference heading towards zero while frontier inference costs rise. Ameya Kanitkar, CTO of Larridin, an AI measurement platform, notes that AI costs were a primary concern six months ago, with companies spending between $20 and $100 per month per LLM subscription. However, as models improved and could handle more complex, agentic tasks, AI services vendors pushed for increased usage, leading to a 10x increase in costs between January and the present. This shift has forced companies to reevaluate their AI spending, with some now allocating 10-20% of their labor costs to tokens, which may not always translate to higher productivity.
Larridin's data reveals an inflection point where burning more tokens fails to boost productivity. By setting a token limit for employees, companies can cut AI costs by 40% without changing anything else. Open weight models offer another cost lever, with Kimi 2.6/2.7 and GLM 5.2 providing parity with Opus4.7 or 4.8 at a fraction of the cost. However, price isn't always the most important consideration. Enterprises still direct almost half of their AI spending toward Anthropic's Opus model due to its superior handling of complex engineering and reasoning tasks.
In conclusion, the AI market is evolving rapidly, with a clear divide between commodity and frontier models. This shift has significant implications for businesses, forcing them to reevaluate their AI spending and consider alternative models. While price is a critical factor, the quality of output and the specific needs of the business must also be taken into account. The future of AI is likely to be characterized by a continued push for innovation and efficiency, with businesses seeking to maximize the value they get from their AI investments.
