OpenAI Decisions API Speeds Up Yes/No Classification Tasks

What happened

OpenAI has launched a Decisions API, a new tool designed to simplify large-scale classification tasks by reducing complex evaluations to a single output: yes or no, a category pick, or a point on a scale. The API processes both text and images and is built specifically for high-volume, repetitive judgement calls rather than open-ended generation.

According to The Decoder, OpenAI says the Decisions API runs roughly ten times faster than its existing Responses API and is priced at $0.10 per million input tokens, positioning it as a lightweight, lower-cost option for developers who need rapid, structured classification rather than full conversational or generative output.

Why it matters

The launch reflects a broader shift in how AI providers are packaging models: not every task needs a general-purpose chat interface, and many enterprise workflows — content moderation, lead scoring, quality triage, compliance checks — are really just repeated binary or categorical decisions at scale. By stripping the task down to a narrow, fast, cheap output format, OpenAI is making it more practical to embed AI judgement directly into high-volume operational pipelines rather than bolting a heavier generative model onto a simple classification problem.

For organisations running digital transformation or automation programmes, this matters because it lowers the cost and latency barrier to using AI for the thousands of small, structured decisions that sit inside customer service, operations and risk workflows — the kind of decisions that are too mundane to justify a full language model call today, but too variable for traditional rules engines.

By the numbers

• 10x faster processing speed claimed for the Decisions API compared with OpenAI's Responses API
• $0.10 per million input tokens, the stated price point for the new API

The Renascence take

The interesting part of this launch isn't the speed or the price — it's the implicit admission that most "AI decisions" inside a business are not creative at all. They are narrow, repeatable classification tasks dressed up as complex judgement calls, and until now they were often routed through tools built for conversation rather than triage.

Most organisations don't actually need an AI that can write an essay about whether a customer complaint is valid — they need a fast, cheap, consistent answer to "yes, no, or which one," applied the same way a thousand times a day. The real behavioral risk isn't the model's accuracy; it's that cheaper, faster classification tools make it easier to automate judgement calls that were quietly relying on human discretion to catch edge cases and exceptions. Teams adopting tools like this should treat the speed gain as licence to redesign the decision itself — tightening definitions, auditing edge cases and setting clear escalation paths — not as permission to remove oversight from decisions that affect real customers.