OpenAI has cut GPT-5.6 Sol API prices by more than 20%, and this isn't news for the sake of news.
Just last week I was talking with a client about the cost of API calls and why it suddenly starts dictating the architecture. When your main vendor cuts prices by that much, you get a reason (and the right) to recalculate everything: where you optimized out of fear of the bill, and where you can bring back quality or simplicity.
What this can mean in practice (no magic, just economics):
1) Recalculate unit cost. How much does one "useful" answer cost: the request, retrieval, the LLM, post-processing.
2) Revisit your optimizations. It may turn out that aggressive caching, trimming context and fighting for every token no longer pay off, or that you can ease up on them.
3) Check the most "expensive" spots. RAG and agent pipelines multiply cost painfully by the number of steps and users.
4) Update budgets and limits. And renegotiate the boundaries with the business: what matters more, speed, quality or price.
Source: Tech in Asia
The image for this post is in the comments.