OpenAI announced that it is fully opening up its Decisions API. This interface is built on GPT-6 Luna, directing the model's smarts specifically at questions that call for "constrained answers" — helping applications and agents make near-real-time judgments based on text, images, and structured data. In other words, it is not built for open-ended creation, but specializes in scenarios where the answer range is clearly defined and quick calls are needed.
The most eye-catching part is how the billing works. The Decisions API charges $0.1 (about 0.67 yuan) per million input tokens, while cache-hit input is pushed down even further to $0.01 (about 0.067 yuan) per million tokens. Crucially, it bills only for input tokens — output tokens are free, so the longer the judgment results, the more users benefit. OpenAI revealed that internal teams are already moving this API into products and business workflows.
Tenfold Speed Boost and Expanded Capabilities
On performance, the company officially says its response speed is up to 10 times faster than calling GPT-6 Luna via the Responses API. Since the public beta, new features have been added one after another: native GPT-Live integration, Prompt caching for reused context, and inference data residency support in the US and the EU.
The freedom given to developers is also expanding. Native JSON input can now be passed directly, images can be embedded in question instructions, and up to 400 questions can be submitted per request — twice the previous limit. Latency performance continues to improve, and service capacity in multiple Asian regions has also been expanded. Together, these changes point to one signal: OpenAI wants to scale Luna's capabilities from "chatting" to "making decisions for you."