OpenAI's Decisions API enters public beta at $0.10 per 1M input tokens, with free output

OpenAI has launched the Decisions API in public beta, an endpoint that reduces evaluations of text, images, or both to a constrained answer: yes/no, a pick from set options, or a rating. Input costs $0.10 per 1M tokens, and output tokens are free, according to the announcement as reported by The Decoder on Oct 7, 2026.

What the Decisions API does

The API supports three response types:

• Yes/no probabilities: a probability-based binary judgment
• Category selection: a pick from predefined options
• Scale-based ratings: a score on a defined scale

OpenAI says the API runs about ten times faster than its Responses API. That figure is an OpenAI claim, and no independent benchmark has been published. The company lists damage detection in photos, automatic routing of customer inquiries, and document classification as target use cases.

The API supports zero-data retention and HIPAA-compliant use in the US and Europe.

Model support and pricing

Item	Detail
Status	Public beta; general availability "coming soon"
Supported model	gpt-6-luna only
Input price	$0.10 per 1M tokens
Output price	Free
Input types	Text, images, or both
Context window	Not yet disclosed
Benchmark scores	Not yet disclosed

No other models are supported at launch. A general availability date has not been announced.

API tier consolidation

OpenAI also cut its paid API tiers from five to three: Build, Launch, and Grow. Organizations move up automatically once their total credit purchases reach the next threshold. The source lists monthly usage limits of $500, $5,000, and $200,000. It does not state the credit-purchase thresholds for each tier, and presumably the limits map to the tiers in the order listed.

Context

The Decoder describes the Decisions API as likely a response to a small trend of "decision models" that an entity identified in the report as Jev started in mid-September. That framing is the outlet's interpretation, not an OpenAI statement.

What this means

The pricing is the notable part. With output tokens free and input at $0.10 per 1M, a classification call costs only what it takes to read the input. That fits workloads like moderation queues, ticket triage, and image screening, where the answer is one token or a probability. Teams currently running these tasks through general-purpose chat or reasoning models with structured output are paying for flexibility they don't use.

The constrained output format probably explains the claimed speed gain. A model that returns a label or a probability avoids generating free text, though OpenAI has not published how the endpoint works internally.

Two caveats apply. First, support for a single model means teams cannot choose a different cost and accuracy tradeoff yet. Second, without published accuracy figures or a context window, buyers have no basis for comparing gpt-6-luna's judgments against their current classifiers. Teams should run their own evaluations on labeled data before migrating.

The tier consolidation is a smaller change. It removes two rungs from the ladder and makes upgrades automatic, but the credit-purchase thresholds still need to be confirmed in OpenAI's documentation.