Model Releases · OCTOBER 24, 2026
OpenAI's Decisions API Goes GA at $0.10 per Million Input Tokens, Output Free
The GPT-6 Luna–powered endpoint moves out of beta on October 22 with 400 questions per request, prompt caching, image URLs and JSON input, and native GPT-Live triggers — targeting a decision-model category where Jev lists at $0.042.
OpenAI moved its Decisions API to general availability on October 22, pricing input at $0.10 per million tokens and charging nothing for output, which quietly reframes what counts as a "model call" inside production systems. The endpoint, powered by GPT-6 Luna, accepts up to 400 questions per request (double the beta ceiling of 200), supports prompt caching above 1,024 tokens, and ships with native GPT-Live triggers alongside image URL and JSON inputs.
The pricing is the story and also isn't. Decisions sits at roughly 2.4x the list price of TypeSafe's Jev, which anchors the category at $0.042. Microsoft Decision-1 posted at $0.042 on October 9, Cloudflare's Clef-flash at $0.038 the same day, Subquadratic SubQ Snap at $0.05 on October 12, and Nace's Drex Max at $0.06 on October 19. OpenAI is the premium tier in a market it didn't create.
What it's selling, instead, is latency and the elimination of output cost. "100, 120, 180 milliseconds," said Nikunj Handa, OpenAI's API product lead, describing a surface where comparable work used to take "one to three seconds." That gap is the entire pitch. Classification, scoring, and routing calls that previously ran through a full chat completion now return a predicate probability between 0 and 1, or a score across two to ten ordered rubric levels, up to 10x faster than the Responses API.
The customer cases cluster around the same observation. Datadog's AJ Stuyvenberg clocked 40x faster p95 latency scoring risk on agent-generated shell commands. Ramp's René Sultan reported 4x faster and 30% cheaper than hitting GPT-6 Luna directly. Sierra's Ajeet Grewal put next-action selection under 280ms p95, 3x faster and 90% cheaper than its previous approach. Speak's CTO Andrew Hsu cut time-to-next-action from 1.4 seconds to 0.13. Box's Biswaroop Palit saw 5x faster route selection on a sample of 30 certificate-of-insurance cases.
The structural read is that OpenAI is doing what hyperscalers have historically done when a cheaper rival defines a category: concede the price axis, win on integration surface. GPT-Live triggers, image URL input, and the Responses-API lineage mean Decisions isn't a standalone product so much as an escape hatch for teams already inside OpenAI's stack. The question is whether a 2.4x premium holds once Jev, Decision-1, and Clef-flash answer back on question-count ceilings and multimodal inputs.
Pricing floors in commodity-adjacent AI tend not to hold for long. The Decisions API is priced like it knows that.
Sources
- https://techcrunch.com/2026/10/22/openai-makes-its-jev-competitor-widely-available-as-decision-model-competition-heats-up/
- https://www.unite.ai/openai-makes-decisions-api-generally-available-with-gpt-6-luna/
- https://cryptobriefing.com/openai-decisions-api-generally-available/
- https://cellcog.ai/blog/openai-decisions-api/
- https://fourweekmba.com/ai-openai-opens-decisions-api-at-2-4x-the-price-of-rival-jev/