OpenAI Opens Decisions API to All, Heating Up Decision-Model Race

OpenAI Opens Decisions API to All, Heating Up Decision-Model Race

OpenAI Pushes Its Decision Model to Every Developer

On October 22, OpenAI announced that its Decisions API is now generally available to all customers, expanding the GPT-6 Luna-based bounded-decision capability from limited preview into a production-ready product. The launch marks OpenAI's formal entry into the fast, cheap decision-model market that TypeSafe's Jev helped create.

Repurposing Luna Into a Decision Engine

OpenAI product lead Nikunj Handa told TechCrunch the team reconfigured ChatGPT Luna to accelerate inference and cut cost. Response times fell from one to three seconds to roughly 100, 120, or 180 milliseconds depending on the request, and OpenAI claims up to 10x faster responses than the Responses API. The API accepts text, images, and structured input, and each request can now carry up to 400 questions, double the public beta limit.

Three Weeks After Debut, Now Fully Open

The Decisions API debuted at OpenAI's September 29 developer day and opened as a public beta on October 6. Ahead of general availability, OpenAI added native GPT-Live integration, prompt caching, JSON input support, and inference in both the US and EU, plus added capacity in multiple Asian regions to further cut latency. At general availability, output stays free and caching moves to Luna's standard rates, with regional processing premiums and long-context multipliers still applying. Handa said customers want the API to be more flexible about using images as outputs, and that Luna is "extremely powerful and has been showing better calibrations" for image classification.

Early Customers Already Replacing Standard Calls

Early customers including Box, Datadog, Ramp, and Speak have already used it to replace standard LLM calls. Internally, OpenAI's dots, Personal Finance, and Communications teams use it to cut response times, classify transaction descriptions, and sort media coverage by topic with sentiment scoring. Pricing is $0.10 per million input tokens with output unbilled, a clear draw for teams making high-volume small decisions. One independent estimate puts a 2-million-token workload at about $0.084 on Jev versus $0.20 on Decisions, though identical text does not guarantee identical billable token totals.

A Head-On Clash With TypeSafe's Jev

TypeSafe reached a $7.6 billion valuation within weeks of launching Jev, which ranks seventh on the AIM System One Index and third on the Artificial Analysis Decision Index; OpenAI's decision model ranks eighth and fourteenth. Jev is cheaper at $0.042 per million input tokens with free output, less than half of OpenAI's rate, but it accepts text only and caps at 255 choices per request, while OpenAI supports image input and holds an edge on compliance credentials (ZDR, HIPAA) and US-EU data residency. The field keeps widening: AWS released the open-source Strands Decider 2B, and Cloudflare launched Clef and Clef-flash. Beyond the benchmarks, the real contest is only beginning.