AIが安くなった——GPT-5.6の衝撃OpenAI announced a major price reduction for GPT-5.6 alongside significant…

OpenAI has announced GPT-5.6, and alongside the new model it disclosed a substantial reduction in the price of using it, framed around an improved price-performance ratio. The move matters because cost has long been one of the main barriers to deploying large language models at scale, and cheaper access could bring capable AI within reach of a wider range of businesses and individual users.

According to the company's official blog post cited by the source, the headline change is not simply a new model but the economics around it. Price-performance ratio refers to how much high-quality output a user can obtain for a given cost; a higher value means better results for the same spend, or comparable results for less money. By emphasizing this metric rather than raw capability alone, OpenAI appears to be signaling that the practical question for many customers is shifting from "can the model do this?" to "can I afford to do this at volume?"

For most commercial language models, usage is billed per token, a unit roughly corresponding to a fragment of a word, with separate rates for input, meaning the prompt, and output, meaning the generated text. Lowering these per-token rates directly reduces the cost of every API call, and that effect compounds quickly for applications that process long documents, run extended conversations, or serve many users. A meaningful cut to token pricing can therefore change which use cases are economically viable, from customer support automation and document summarization to code generation and data extraction.

The announcement fits a broader pattern across the industry, in which the cost of a given level of model performance has fallen steadily over the past few years. OpenAI has previously introduced lower-cost tiers and smaller, faster variants of its models, and competitors have done the same. Anthropic offers its Claude family across multiple size and price points, and Google positions its Gemini models similarly, with each provider offering cheaper options intended for high-volume or latency-sensitive workloads. Continued price competition among these vendors is likely one of the factors driving the trend.

Several forces appear to be pushing prices down. Improvements in model architecture and training efficiency, more capable inference hardware, and techniques such as distillation, quantization, and better serving infrastructure all reduce the cost of running a model. As providers gain scale, they can also pass some of those savings to customers while competing for developer mindshare. The exact contribution of each factor to GPT-5.6's pricing is not detailed in the source, so specifics should be treated with caution.

For organizations evaluating the change, the practical takeaway is that lower prices can expand the set of workloads worth automating, but total cost still depends on how a system is designed. Prompt length, the amount of generated output, retrieval and caching strategies, and whether an application calls the model once or repeatedly all shape real-world spending. Buyers comparing providers typically also weigh factors beyond headline price, including latency, rate limits, context window size, reliability, data-handling policies, and how well a model performs on their specific tasks.

It is worth noting that the reporting here originates from a blog post rather than a first-party product page, and the excerpt does not enumerate the specific new per-token rates. Readers who need to plan budgets should confirm the current figures on OpenAI's official pricing documentation, since published rates and available tiers can change and may vary by region, model variant, and usage volume.

Still, the direction is clear enough. If the improved price-performance ratio holds up under real workloads, GPT-5.6 could accelerate a trend that has already been reshaping how teams think about building with AI: treating advanced language models less as a premium, carefully rationed resource and more as a routine part of everyday software. Whether this particular release proves to be a turning point or simply the next step in a longer decline in AI costs will become clearer as developers test it against their own applications and publish independent comparisons.