OpenAI launches Decisions API with image input for developers

By Splitfeed EditorialLast updated Oct 7, 2026, 15:40 UTC

On October 7, 2026, OpenAI made its Decisions API available to all developers, pairing the launch with image input support and a significant price point of $0.10 per million input tokens. The API is built around structured outputs, returning yes/no probabilities, category selections, or scale ratings rather than open-ended text. OpenAI says it processes requests roughly ten times faster than its existing Responses API. Alongside the release, the company trimmed its paid API tiers from five to three, consolidating its developer-facing commercial structure.

The launch matters for a few reasons beyond the speed and pricing figures. Classification at scale is one of the more common and cost-sensitive tasks developers actually build with, covering everything from content moderation to document routing to automated decision pipelines. By bundling image input into the same structured-output framework, OpenAI is extending that capability beyond text, which puts it into more direct competition with TypeSafe's Jev product in a segment that has so far been less dominated by general-purpose language models. The tier reduction from five to three also signals a deliberate effort to simplify the developer experience, though whether that simplification benefits or constrains different classes of users is a genuine open question.

None of the three camps have published reactions to this story yet. The Pro-AI camp would typically highlight the speed and cost efficiency as evidence that capable AI tools are becoming more accessible to developers of all sizes. The Anti-AI camp would likely raise concerns about making automated classification systems, including image-based ones, faster and cheaper to deploy at scale, particularly where those systems inform real-world decisions about people. The Middle Ground camp would probably focus on the practical tradeoffs, welcoming the infrastructure improvements while calling for clear documentation of the error rates and limitations that come with reducing complex evaluations to binary or categorical outputs.

The most concrete things to watch are how developers report real-world accuracy compared to the speed claims, whether TypeSafe responds with updates to Jev, and how the revised three-tier pricing structure settles for teams that were previously on the tiers that no longer exist.