OpenAI's Decisions API goes generally available: $0.10 per million input tokens, zero output fees

#What happened

OpenAI has opened its Decisions API to everyone with API access. General availability arrived on October 22, 2026, three weeks after the public beta that began October 6 and a little over three weeks after the limited preview at DevDay on September 29. The API is designed for high-volume, low-latency decision-making: instead of returning paragraphs, it produces structured, typed answers in the form of probabilities, choices, and scores. It runs on GPT-6 Luna and targets routing requests, classifying inputs, and repeated micro-decisions. Pricing is $0.10 per million input tokens; output costs nothing, and neither do cache reads or cache writes. One request can now carry up to 400 questions, a ceiling raised from the beta period. OpenAI also shipped several upgrades between beta and general availability: native GPT-Live integration, prompt caching, and support for image and JSON inputs. On compliance, the API supports Zero Data Retention and is HIPAA eligible, positioned for customers in both the US and European markets. OpenAI reports the Decisions API runs up to 10 times faster than using GPT-6 Luna through the standard Responses API for bounded decisions, with demonstration latency around 150 milliseconds.

Verified October 22, 2026.

#Why it matters

This is OpenAI's answer to the "stop paying for words you parse" problem — by charging only for input tokens and returning a typed verdict, the cost of a call becomes predictable from what a developer sends in. For classification, routing, and micro-decision workloads at scale, the zero-output-fee model removes the bill for generated text that code would immediately discard. The 400-questions-per-request capacity plus image and JSON input support makes it a serious option for high-throughput pipelines, though the 10x speed figure is framed as an upper bound from demonstrations, not an independent benchmark.