The artificial intelligence market is fracturing into two distinct tiers, with budget-friendly models becoming commodities while cutting-edge reasoning systems command premium prices that are climbing faster than ever. This split is forcing companies to rethink how they budget for AI, because spending more on tokens doesn't automatically translate to better results.

Why Are AI Prices Moving in Opposite Directions?

The pricing divergence reflects a fundamental shift in how AI is being used. Standard inference, the process of running a trained model to generate outputs, has become a commodity business. In late 2022, a GPT-4-class model cost about $20 per million tokens of output. Today, equivalent capability costs roughly $0.40, a 55-fold decline in less than four years. When DeepSeek released its R1 reasoning model in January 2025 at $0.55 per million input tokens and $2.19 for output, compared to OpenAI's o1-preview at $15 and $60, the market repriced overnight with a 97 percent discount.

Meanwhile, frontier models, those designed for complex reasoning and specialized tasks, are moving in the opposite direction. OpenAI doubled the price of GPT-5.5 to $5 input and $30 output per million tokens. Google's Gemini Flash 3.5 arrived three to six times more expensive than the model it replaced. Anthropic's Claude Sonnet 5, despite having a lower per-token price than Claude Opus 4.8, uses more tokens to produce the same results, effectively raising the total cost.

How Are Companies Actually Spending on AI Now?

The real shock isn't the per-token price; it's the total bill. Six months ago, companies spent between $20 and $100 per month per large language model (LLM) subscription. Around February 2026, vendors began pushing for higher usage as models improved and could handle more complex agentic work, which takes longer to complete. The result: costs jumped roughly 10-fold between January and now, particularly in engineering operations.

Ameya Kanitkar, Chief Technology Officer of Larridin, an AI measurement platform, explained the disconnect between spending and productivity. "On average we have seen the cost go up about 10x between January and now, especially in engineering ops," he said in an interview. The shift toward longer, agentic tasks and metered pricing models is driving the increase.

Ameya Kanitkar, Chief Technology Officer of Larridin, an AI measurement platform

"The new trend that is emerging is that the open source models, open weight models are actually not that far behind the frontier models. And now the costs are hitting the balance sheet, which are not the real costs, companies have started truly thinking, 'okay, how can we actually adjust these costs?'" said Kanitkar.

Ameya Kanitkar, Chief Technology Officer at Larridin

Larridin's data reveals a troubling pattern: companies are spending between 10 and 20 percent of their labor costs on AI tokens. For a software engineer earning $200,000 annually, that translates to $2,000 to $4,000 per month in token expenses. Yet higher spending doesn't guarantee higher productivity.

Where Does Productivity Actually Peak?

Larridin identified a critical inflection point in its analysis of token spending versus developer productivity. Between 15 and 30 percent of AI users among the platform's clients account for more than 50 percent of total AI spend, and often that spending does not correlate with gains in output. When the company plotted token spend against developer productivity, it found that burning more tokens failed to boost productivity beyond 35 to 40 percent of client spending.

This discovery has practical implications. Using that inflection point as a token limit for employees can cut AI costs by 40 percent without changing anything else, according to Kanitkar. The finding suggests that companies are not optimizing their AI usage; they are simply spending more without strategic oversight.

How to Optimize AI Spending Without Sacrificing Quality

• Implement Token Limits: Set spending caps at the 35 to 40 percent inflection point where additional tokens stop improving productivity, potentially reducing costs by 40 percent without performance loss.
• Use Multiple Models Strategically: Nearly 75 percent of companies now use multiple models, switching between them based on task requirements; this approach is most viable for software development but more difficult for customer-facing agentic work.
• Evaluate Open-Weight Alternatives: Models like Kimi 2.6/2.7 and GLM 5.2 are nearly at parity with Opus 4.7 or 4.8 and cost 10 times less in theory, or about 5 times less in practice, though they may consume more tokens and run slower.

Open-weight models, those released with publicly available parameters, offer another cost lever. Kimi 2.6/2.7 and GLM 5.2 are nearly at parity with Anthropic's Opus 4.7 or 4.8 and are 10 times cheaper in theory, or about 5 times cheaper in practice. They tend to be slower and consume more tokens on a per-token basis, but the overall token cost remains low.

Switching between models is more difficult for customer-facing agentic work, where consistency and reliability matter most. For internal software development, however, it is much more viable. Enterprises still direct almost half of their AI spending toward Anthropic's Opus model because it handles complex engineering and reasoning tasks exceptionally well, suggesting that price is not always the primary consideration.

What Does This Mean for the Future of AI Budgeting?

The market is clearly splitting into two tiers: commodity inference heading toward zero cost, and frontier inference with rising prices. This bifurcation forces companies to make strategic choices about where to invest their AI budgets. The days of treating all AI spending as a simple line item are over. Organizations that measure token efficiency, set spending limits based on productivity data, and strategically mix models across tasks will emerge with lower costs and better outcomes. Those that simply chase the latest frontier model without understanding their actual productivity gains will find their AI budgets spiraling out of control.