OpenAI cuts GPT-5.6 prices by up to 80% as cheap open-source models close the intelligence gap

OpenAI announced that GPT-5.6 Sol helped make itself more efficient to run. Shortly afterward, the company reduced the price of GPT-5.6 Luna by 80% and GPT-5.6 Terra by 20%, just three weeks after launching the models.

The growing focus on improving token efficiency appears to be driven by companies becoming increasingly sensitive to AI costs. As a result, inexpensive Chinese open-source models are taking center stage and becoming popular alternatives to proprietary American models.

From my perspective, the rise of token efficiency has gradually changed the software development paradigm. Developers are moving away from simply using models to solve individual tasks and toward designing agents that can independently pick up and complete those tasks.