OpenAI Decisions API Turns Prompts Into Typed Branching Signals for Product Code
OpenAI’s public beta Decisions API returns probabilities, fixed choices, and rubric scores from gpt 6 luna inputs, targeting routing and classification loops that no longer need JSON parsing.
What changed
OpenAI released the Decisions API in public beta on 6 October 2026, giving developers a dedicated POST /v1/decisions endpoint backed by gpt-6-luna. Instead of generating free text that application code must parse, the API returns typed outputs: predicate probabilities, choices from fixed option lists with confidence scores, or numeric rubric scores for text and image inputs.
OpenAI states the endpoint is about ten times faster than using GPT 6 Luna through the Responses API for the same classification pattern, and prices input at 0.10 dollars per million tokens with no output token charges on Decisions calls. SDK support landed in the same week across Python, JavaScript, Go, Ruby, and Java clients.
Why it matters
Most production agents spend a surprising fraction of latency budget asking tiny questions: which tool, which model, which escalation tier. A first class decisions endpoint is an architectural nudge toward treating those gates as inference primitives with calibrated probabilities rather than brittle JSON prompts.
Platform teams should still validate calibration on their own data; OpenAI’s docs emphasize tuning thresholds with labeled examples rather than trusting default scores.
Who is affected
Applied AI engineers building routers, trust and safety classifiers, customer support triage, and multimodal inspection pipelines should pilot Decisions on one high volume branch point and compare end to end latency and error rates against Structured Outputs on Responses.
FinOps owners should model spend under input only pricing, especially if image heavy predicates dominate traffic.
What to do next
Pick one production branch with a fixed option set, implement parallel calls to Decisions and your existing Responses path for a week, and log disagreement cases. Only migrate traffic if typed outputs reduce parser failures without increasing false escalations.
What to watch
Watch for general availability timing, additional model support beyond gpt 6 luna, and published calibration benchmarks from OpenAI or independent evaluators.
Sources
- Primary. OpenAI Developer Community, Decisions API is now available in Public Beta (6 October 2026). Launch date, model support, output types, and pricing summary.
- Primary. OpenAI documentation mirror, Decisions guide snapshot (6 October 2026). Endpoint semantics for predicate, choice, and score questions.
- Secondary. MarkTechPost, OpenAI Decisions API Hits Public Beta With 10x Faster Typed Answers (9 October 2026). Independent summary of developer use cases and comparison framing.