OpenAI has opened its Decisions API to all customers, entering a market for small, low-cost AI models that make split-second choices instead of generating long text. The API moved to general availability on October 22, 2026, adding voice and visual modalities and routing inference geographically to lower latency worldwide.
A decision model works like a super-fast sorter: hand it a question and a set of predefined options, and it quickly picks one with a confidence score. That makes it ideal for routing tickets, flagging content, or deciding an agent’s next step. Early trials with customers including Box, Datadog, Ramp, and Speak replaced LLM calls with the API to boost throughput; Speak says it cut the lag between actions from 1.4 seconds to roughly 0.13 seconds.
OpenAI did not train a purpose-built model for this. The team took GPT-6 Luna, the model behind ChatGPT‘s everyday experience, and reconfigured it to run faster and cheaper. Product lead Nikunj Handa says the big unlock is speed: response times fell from one to three seconds down to roughly 100 to 180 milliseconds, about ten times faster than the same model through OpenAI’s Responses API. Input pricing starts at $0.10 per million tokens, with no output charge.
The category was effectively created by startup TypeSafe and its Jev model, which turned the company into a $7.6 billion business within weeks. Competition is already fierce: public benchmarks such as the AIM System One Index favor Liquid’s D1 Ultra and Base Lab’s Elephant-1, and OpenAI currently ranks mid-pack on those indices. OpenAI’s argued advantage is multimodality — Luna shows better calibration on image classification, and customers increasingly want decisions that include images. It enters a crowded field alongside Google’s Gemini and Anthropic’s Claude as developers weigh which model family powers their next build.
Source: Zotpaper
