OpenAI adds Decisions API for yes/no, category and rating tasks

OpenAI has announced a new Decisions API for evaluating text, images, or both, according to The Decoder. The company says it is designed for outputs such as yes/no probabilities, choosing from preset categories, or assigning ratings, which matters for teams trying to turn AI judgments into automation steps rather than long-form responses.

The launch also comes with a cleanup of OpenAI’s paid API tiers, reducing them from five to three. For creators, marketers, freelancers and small businesses building lightweight AI systems, the update points to a more specialized way to handle classification and routing tasks.

What the Decisions API is built to do

The Decisions API is in public beta now, with general availability described as coming soon. OpenAI says the API can evaluate text, images, or a combination of the two.

Unlike a general chat-style response, the API focuses on constrained decisions. The supported response types are:

• yes/no probabilities
• selection from predefined categories
• scale-based ratings

OpenAI says possible uses include detecting damage in photos, routing customer inquiries automatically, and classifying documents. Those examples are important because they are everyday business problems: deciding what something is, where it should go, or whether it meets a condition.

The company says the new API runs about ten times faster than the Responses API. It currently supports only gpt-6-luna, according to the report. Pricing is listed at $0.10 per million input tokens, and output tokens are free.

Why constrained answers matter for automations

For small teams, the practical appeal is not that the model writes more. It is that it may write less, in a format that is easier to act on.

A yes/no probability, a category pick, or a rating can be mapped into a business rule more cleanly than a paragraph. A support request can be routed to a queue, a document can be tagged, or an image can be flagged for review. That is the kind of output that fits naturally into workflow tools, spreadsheets, dashboards, and internal apps.

This does not mean the Decisions API replaces workflow platforms such as Zapier, Make, or n8n. The source does not mention native connectors or integrations. But for builders already using APIs inside automations, a narrower decision-focused endpoint could reduce the amount of prompt handling and result parsing needed around classification tasks.

Where it sits beside ChatGPT and general AI APIs

The Decisions API is part of OpenAI’s API platform, not a new consumer feature inside ChatGPT. That distinction matters for non-technical readers: ChatGPT is typically used through a chat interface, while APIs are used by developers and product teams to add AI behavior inside software, workflows, or internal systems.

For teams currently using broader AI endpoints to classify messages, score leads, review images, or label documents, OpenAI is positioning this as a more specialized alternative. The Decoder also notes that the move appears to fit a recent trend around “decision models,” following activity from Jev in mid-September.

OpenAI also says the Decisions API supports zero-data retention and HIPAA-compliant use in the US and Europe. For organizations dealing with sensitive workflows, those support details may be as important as speed or price, though teams will still need to assess their own compliance requirements.

API tiers are being simplified

Alongside the new API, OpenAI has reduced its paid API tiers from five to three. The new tiers are called Build, Launch, and Grow.

Organizations move between tiers automatically when their total credit purchases reach the next threshold, according to the report. Monthly usage limits are $500 for Build, $5,000 for Launch, and $200,000 for Grow.

For freelancers and small businesses, the practical takeaway is that OpenAI is trying to make its API platform easier to understand while also adding narrower tools for production workflows. The Decisions API will be most relevant to teams that repeatedly ask AI to make the same type of judgment, especially where the answer needs to become an action in another system.