OpenAI Introduces Decisions API to Simplify Complex Evaluations into Yes, No, or Choose One

Meta description: OpenAI Decisions API public beta runs about ten times faster than Responses API, while paid API tiers shrink from five to three under new limits.

Tags: OpenAI, Decisions API, gpt-6-luna, API pricing, AI infrastructure

Featured image: Abstract technology style network of decision nodes, no real people, no company logos, no text.

Background: From Broad Chat APIs To Structured Decisions

OpenAI announced two updates to its API platform: the OpenAI Decisions API, now in public beta, and a reduction of paid API tiers from five down to three. The pair of changes reflects OpenAI's effort to expose specialized endpoints while simplifying how customers purchase API capacity. The Decisions API is likely OpenAI's response to the mini-trend of decision models that Jev kicked off in mid-September. In that context, a decision model is usually treated as a narrower tool than a chat model: it is expected to return a bounded answer, such as a category, a probability, or a rating, rather than a long response.

This background matters because many software teams need AI output that can be wired directly into workflows. A routing decision, a classification label, or a yes/no risk check is easier to validate than free-form text. The new API's focus on text, images, or both makes it relevant for visual inspection and document processing, while its beta status means OpenAI is exposing the endpoint before general availability, allowing developers to test behavior, cost, and latency.

Development: What The Decisions API Does And How It Is Priced

The Decisions API evaluates text, images, or both, and OpenAI says it runs about ten times faster than the Responses API. Speed is the headline number because the endpoint is positioned for automated decisions, where latency can affect user experience and workflow throughput. Only gpt-6-luna is supported so far, priced at $0.10 per million input tokens. Output tokens are free, which shifts the economic emphasis toward controlling prompt size and input content rather than generating long outputs.

The API handles three response types: yes/no probabilities, picks from predefined categories, and scale-based ratings. These formats map directly to common automation tasks. OpenAI lists damage detection in photos, automatic customer inquiry routing, and document classification as use cases. The privacy and compliance features are also part of the pitch: it supports zero-data retention and HIPAA-compliant use in the US and Europe. Zero-data retention is a privacy posture aimed at avoiding storage of API content for product improvement, while HIPAA refers to healthcare information handling rules in the US.

Latest Progress: Three Tiers, Beta Access, And What Comes Next

OpenAI also cut its paid API tiers from five down to three: Build, Launch, and Grow. Organizations move up automatically when their total credit purchases hit the next threshold. Monthly usage limits sit at $500, $5,000, and $200,000. The tier ladder is simpler than the previous five-level structure, but it still uses a predictable upgrade trigger: cumulative credit purchases. For teams evaluating the platform, this creates a clearer path from low-volume testing to higher-volume production use.

The Decisions API is in public beta now, with general availability coming soon. That places the product in a transitional stage: it is available for testing, but support is limited to gpt-6-luna, and the pricing model is still tied to early access. The key numbers are the ten times speed claim, the $0.10 per million input token price, the free output tokens, the three response types, and the $500, $5,000, and $200,000 monthly usage limits across the new tiers.

The current stage is early availability. The API is live enough for evaluation, but it remains beta-only and model-limited, while the tier change gives customers a simpler purchasing path. What to watch next is general availability, support for additional models, and whether the faster, decision-oriented endpoint becomes a standard part of OpenAI's developer stack.