OpenAI Decisions API launched in beta on 6 Oct and went GA on 22 Oct 2026. It costs $0.10 per million uncached input tokens, with free output.
By OpenAI · Public beta launch: 6 Oct 2026
OpenAI's Decisions API uses GPT-6 Luna to return typed answers for classification, routing and agent actions. It launched in public beta on 6 Oct 2026 and became generally available on 22 Oct. Developers supply the questions and allowed answers; the endpoint returns probabilities, choices or rubric scores.
OpenAI entered the decision model market with the public beta launch on 6 Oct 2026. Native GPT-Live integration followed on 15 Oct. The API became generally available on 22 Oct, doubling the per-request question limit from 200 to 400.
The published rate is $0.10 per million uncached input tokens, with free output. Prompt-cache reads and writes use standard GPT-6 Luna caching rates; regional and long-context input multipliers apply.
POST /v1/decisions accepts gpt-6-luna only. Predicate questions return a probability; choice questions select from allowed options; score questions return a probability-weighted level index. Inputs can include text, images, JSON and conversation history with tool calls and outputs. Files, audio and item references are unsupported.
Parallel's own PWS-Decisions benchmark gives OpenAI 41.67%, against 76.22% for Parallel Decisions API with live search. OpenAI had no web access or supplied external context, so the setup tests different access to evidence. These are VENDOR-REPORTED results from Parallel. In XDecideBench, GPT-6 Luna Decisions ranks eighth of ten complete Core12 runs but first on native INCLUDE exams at 0.600. SILT's calibration result for OpenAI is inconclusive. These tests cover different tasks and scoring rules.
PWS-Decisions evidence · XDecideBench evidence · SILT calibration evidence · Compare decision models · Jev alternatives
Source: Decisions guide → Source: API changelog →
A live site where OpenAI Decisions, Jev and Cloudflare's Clef put a probability on live questions after reading recent news, never the odds. Their probabilities are compared against Polymarket and Kalshi prices.
A proof of concept that classifies incoming requests with the Decisions API preview and hands them to worker agents. Its README warns that the preview's request format is not yet published.
Calls Jev, OpenAI's Decisions API, Upstage Solar Decide and local Ollama models through one interface and logs version, latency, tokens and cost. On picking the right one of 12 tools for 40 Korean and English requests, the maker measured Jev 1.13 at 40 of 40 and OpenAI Decisions at 0.90, against 0.85 for a generative model choosing directly.