{"slug": "openai-decisions-api-is-ga-10x-faster-ai-routing", "title": "OpenAI Decisions API Is GA: 10x Faster AI Routing", "summary": "OpenAI made its Decisions API generally available on October 22, offering a POST /v1/decisions endpoint that returns typed judgments — predicate probabilities, choices from a developer-defined list, or scores on an ordered scale — up to 10 times faster than the standard Responses API for classification workloads, with no output-token charges. The GA release adds native GPT-Live integration for voice agents, native JSON inputs, US and EU inference residency, and a doubled per-request question limit from 200 to 400, at $0.10 per million input tokens ($0.01 with prompt caching on prefixes of 1,024 tokens or more). OpenAI's pricing is 2.4 times that of Jev, the TypeSafe AI product that popularized the bounded-answer model about two weeks earlier at roughly $0.04 per million input tokens, and the API does not stream or support function calling, conversation state, or custom JSON schemas.", "body_md": "OpenAI’s Decisions API went generally available on October 22, formalizing an endpoint that returns typed judgments instead of prose. Feed it context — text, images, or JSON — alongside questions that carry a fixed set of legal answers, and it comes back with a structured result up to 10 times faster than the standard Responses API, with no output-token charges on top.\n\n## Three Question Types, One Endpoint\n\nThe API lives at `POST /v1/decisions` and supports three question shapes. A **predicate** returns a probability that a condition is true. A **choice** picks one value from a list you define. A **score** returns a position on an ordered scale. All three return typed answers — not a paragraph you parse yourself.\n\nA support-ticket router illustrates the pattern:\n\n``` python\nfrom openai import OpenAI\nopenai = OpenAI()\n\ndecision = openai.decisions.create(\n    model=\"gpt-6-luna\",\n    input=\"I was charged twice for my order.\",\n    questions=[{\n        \"type\": \"choice\",\n        \"name\": \"department\",\n        \"instructions\": \"Which department should handle this?\",\n        \"choices\": [\n            {\"value\": \"billing\",   \"description\": \"Payments and refunds.\"},\n            {\"value\": \"technical\", \"description\": \"Product bugs.\"},\n            {\"value\": \"shipping\",  \"description\": \"Delivery issues.\"},\n        ]\n    }],\n)\nprint(decision.answers[0].choice)  # \"billing\"\n```\n\nYou can ask up to 400 questions per request and pass up to 128 images. The endpoint handles the rest.\n\n## Speed and What It Costs\n\nOpenAI claims 10x faster throughput versus the Responses API for classification workloads. Pricing at GA is $0.10 per million input tokens with zero output-token charges — because there are no output tokens, just typed fields. Prompt caching drops that to $0.01 per million tokens for repeated prefixes of 1,024 tokens or more.\n\nThat pricing deserves scrutiny. Jev, the TypeSafe AI product that popularized the bounded-answer model concept about two weeks before OpenAI shipped this, charges roughly $0.04 per million input tokens — making OpenAI’s offering 2.4 times more expensive. The open-source ecosystem moved fast too: within four weeks of Jev’s launch, [more than 20 alternatives appeared](https://www.marktechpost.com/2026/10/17/20-jev-alternatives-the-decision-models-taking-on-typesafe-in-2026/), including Cloudflare’s Clef, Jared Palmer’s Kev (Qwen 3-based), and Foq, a local option clocking in around 25 milliseconds. OpenAI’s brand carries weight, but cost-sensitive workloads have real options.\n\n## What GA Added\n\nThe public beta ran lean. The GA release layered on four meaningful additions. Native [GPT-Live integration](https://developers.openai.com/api/docs/guides/decisions) means voice agents can make structured judgments mid-conversation without breaking out to a separate API call. That is not a minor detail: Speak, the language learning app, reported cutting its time-to-next-action from 1.4 seconds to 0.13 seconds after switching — a reduction of more than 90 percent. Native JSON inputs let developers pass structured app data directly instead of serializing it into a string. US and EU inference residency was added for data sovereignty requirements. And the per-request question limit doubled from 200 to 400.\n\n## What This Does Not Replace\n\nThe bounded-answer constraint is a feature, not a limitation — but it means the Decisions API is the wrong tool for roughly half of what developers reach for the Responses API to do. It does not stream. It does not support function calling, conversation state, or custom JSON schemas. The developer community’s reaction at launch was measured; one Hacker News commenter cut to it: “It’s not guaranteed to be correct: it’s guaranteed to be *formatted in a particular way*. You can get the same thing with grammars on any LLM.”\n\nThat criticism is accurate but also misses the point. Yes, you can replicate this pattern with structured outputs and a general-purpose model. The Decisions API makes it faster to implement, faster at runtime, and simpler to maintain — which matters at scale.\n\n## When to Reach for It\n\nThe Decisions API earns its place in pipelines where every decision node has a known answer space: routing an inbound request, classifying content before storage, choosing which tool an agent should call next, or gating content against a rubric. If the answer could be anything — if you need generation, not judgment — stay on the Responses API. If the answer is always one of five things, this endpoint is the right tool, and the pricing gap with Jev is worth weighing against the integration simplicity of staying in [OpenAI’s ecosystem](https://www.unite.ai/openai-makes-decisions-api-generally-available-with-gpt-6-luna/).\n\n*The [full API reference](https://developers.openai.com/api/docs/guides/decisions) is available in OpenAI’s developer documentation.*", "url": "https://wpnews.pro/news/openai-decisions-api-is-ga-10x-faster-ai-routing", "canonical_source": "https://byteiota.com/openai-decisions-api-ga-gpt6-luna/", "published_at": "2026-10-26 04:09:03+00:00", "updated_at": "2026-10-26 04:15:16.682006+00:00", "lang": "en", "topics": ["ai-products", "developer-tools", "large-language-models", "ai-agents"], "entities": ["OpenAI", "Decisions API", "Responses API", "GPT-Live", "Jev", "TypeSafe AI", "Cloudflare", "Speak"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/openai-decisions-api-is-ga-10x-faster-ai-routing", "markdown": "https://wpnews.pro/news/openai-decisions-api-is-ga-10x-faster-ai-routing.md", "text": "https://wpnews.pro/news/openai-decisions-api-is-ga-10x-faster-ai-routing.txt", "jsonld": "https://wpnews.pro/news/openai-decisions-api-is-ga-10x-faster-ai-routing.jsonld"}}