OpenAI's GPT-5.5 pricing shows a stark disconnect between headline rates and real-world costs. While OpenAI doubled the list price compared to GPT-5.4, claiming shorter responses would compensate users, actual usage data paints a different picture. An OpenRouter analysis of real deployment data found that costs increased 49 to 92 percent depending on input length, directly contradicting OpenAI's efficiency claims.
The pricing structure hits users differently based on their workload. Short-context applications see smaller cost increases around 49 percent, while longer-context queries face steeper jumps approaching 92 percent. This variation matters for enterprises running mixed workloads, where some applications benefit from longer context windows while others don't.
OpenAI isn't alone in this trend. Anthropic recently raised prices on Claude Opus 4.7, signaling industry-wide cost increases. Both companies face pressure to improve margins as they approach potential IPOs. Public markets reward revenue growth and profitability, creating incentive to monetize model improvements more aggressively rather than passing efficiency gains to customers.
The broader pattern reflects a maturing AI market moving away from aggressive pricing to gain users. Early large language model pricing emphasized volume and market share. Now that major models dominate their segments, vendors extract higher margins on their core offerings. OpenAI's messaging about efficiency improvements rings hollow when actual costs rise substantially.
For enterprise customers, the increases force recalculation of AI budgets and model selection strategies. Some may switch to cheaper alternatives or reconsider context window usage to control costs. Developers building atop these APIs face margin compression unless they can raise prices on their own services.
The trend also reveals how list prices increasingly diverge from practical costs in the AI market. Vendors adjust per-token rates, output token multipliers, and batch pricing separately, making true