--- title: OpenAI's Decisions API is a sub-second classifier, and... description: OpenAI has released a Decisions API in limited preview, powered by a specialized model called Luna (part of GPT-6), designed to pick from a defined set of... canonical: https://daily.dev/posts/openai-s-decisions-api-is-a-sub-second-classifier-and-people-are-already-calling-it-a-jev-killer-ldvisly4o twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: OpenAI's Decisions API is a sub-second classifier, and people are already calling it a Jev killer | daily.dev og:description: OpenAI has released a Decisions API in limited preview, powered by a specialized model called Luna (part of GPT-6), designed to pick from a defined set of... og:url: https://daily.dev/posts/openai-s-decisions-api-is-a-sub-second-classifier-and-people-are-already-calling-it-a-jev-killer-ldvisly4o og:image: https://api.daily.dev/og/posts/lDVIslY4O.png og:image:alt: OpenAI's Decisions API is a sub-second classifier, and people are already calling it a Jev killer og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # OpenAI's Decisions API is a sub-second classifier, and people are already calling it a Jev killer **[Trends](https://daily.dev/sources/trends)** · 3 min read · 2 upvotes · 0 comments ## Summary OpenAI has released a Decisions API in limited preview, powered by a specialized model called Luna (part of GPT-6), designed to pick from a defined set of answers in under a few hundred milliseconds, including visual inputs. Positioned for content classification, request routing, and agent decision-making, it's being compared to Jev, an existing constrained decision-making tool, with some already calling it a 'Jev killer.' The comparison and hype are running well ahead of any public benchmarks since access remains limited. ## Content OpenAI used DevDay to announce the Decisions API, and the timing is hard to miss. TypeSafe AI had released Jev, a fast classifier-style decision model, only weeks earlier, and developers were buzzing about it. The Decisions API does nearly the same job. ## What it does The API runs on OpenAI's small Luna model. Instead of chat-style prose, it picks from answers you define in advance and returns them with confidence scores. OpenAI lists three uses: classifying content, routing requests, and choosing an agent's next action. In the company's own words, you define questions and possible answers, and the model decides. Speed is the selling point. @thsottiaux describes it as lightning-fast constrained decision making that accepts visual inputs and is tuned to respond in a few hundred milliseconds end to end. OpenAI's headline number is 150ms, against 1.6 seconds for GPT-6 Luna used as a general model. It's in limited preview, with a broader rollout promised soon. Pricing, candidate limits and fine-tuning options haven't been disclosed. ## How fast it shipped On a DevDay podcast episode, OpenAI's Ari Weinstein (Computer Use lead, ex-Sky cofounder) and Nikunj Handa (API product lead) said the team built the Decisions API in about a week, on existing Luna weights, after the developer excitement around Jev. That speed explains why some people are calling it a clone. @iannuttall's first reaction was simply: "Jev killer?" ## TypeSafe's view TypeSafe CEO Diogo Almeida joked about "clone wars," but he also read OpenAI's move as validation. His argument is that many software tasks don't need a full LLM. A cheap, fast classifier-style model, what he calls a "System One" approach, handles them fine. One use case he points to is watching AI agents for bad actions. A hackathon demo showed Jev checking agentic actions for $2.94, where a frontier LLM would have cost $372. A monitor that cheap could catch incidents like the recent Hugging Face agent mishap before they spread. ## The rest of the DevDay lineup The Decisions API was one announcement among several, all discussed on the same podcast: - **Dots:** a personal assistant with its own cloud Linux computer. - **GPT-6.1 Sol:** a cheaper, faster model aimed at Computer Use. - **Computer Use in the Agents API.** - **App Shots:** richer context than plain screenshots. The Computer Use conversation was the most interesting part to me. Agents now write and execute JavaScript, combine screenshots with accessibility trees and DOM data, debug their own failures, and recover. Weinstein and Handa say they're approaching superhuman speed on software tasks. They also covered async tool calling, mid-turn steering, WebSockets, UltraFast inference, prompt caching and pre-warming, and context compaction for long-running agent threads. ## What to watch I can't tell yet whether OpenAI matched Jev or overtook it. Without pricing or limits, the 150ms figure is half the story. But it's telling that a frontier lab looked at a small, cheap, single-purpose model and built its own version in a week. If fast constrained decisions become a standard API primitive, the open question is who owns that layer: the frontier labs, or specialists like TypeSafe. ## Questions this post answers ### What is OpenAI's Decisions API and what is it used for? It is a narrow, latency-optimized API in limited preview that lets developers define a question and a set of possible answers, with a specialized model called Luna (part of GPT-6) picking one in under a few hundred milliseconds, including from visual inputs. It targets content classification, request routing, and choosing an agent's next action rather than general-purpose reasoning. _Anyone weighing constrained decision tools for agent routing can follow how this API develops on daily.dev._ ## Community take How the wider developer community reacted, aggregated from 3 discussions and 501 comments across x (as of 2026-09-30). **TL;DR:** Reception is dominated by comparisons to Jev, with many arguing Decisions API is a fast follow or thin wrapper around Luna rather than something novel, while others note real advantages like multimodal (image) support and OpenAI's platform reach. **Sentiment:** 20% positive · 35% mixed · 45% skeptical **The case for** - Adds image/multimodal input support that Jev reportedly lacks. - Being on an established platform with broad developer reach could matter even without technical superiority. - Some testers report Luna as more accurate than Jev, even if slower. - No fine-tuning or training required to use it, similar to Jev's ease of use. **The pushback** - Many see it as just a Luna wrapper doing what structured-output prompting already accomplished since GPT-3. - Several say Jev is still faster and more accurate in their own testing. - Pricing is unannounced, so cost competitiveness versus Jev is unknown. - Some find the image-input use case for yes/no decisions niche or unconvincing. - Perceived as a rushed, copycat reaction alongside other DevDay announcements mirroring competitors. **By community** - x (mixed): Replies swing between dismissing it as a rebranded Luna prompt and acknowledging real speed/multimodal gains, with constant head-to-head comparisons to Jev. **Hottest debate:** Whether Decisions API offers genuine technical advantages over Jev or is just a marketing repackaging of existing Luna prompting techniques. **Open questions** - What will pricing be, and will it undercut or match Jev's cost? - How does it actually benchmark against Jev in real accuracy/speed tests? - Will fine-tuning or candidate-answer limits be supported? - When will general availability open up beyond limited preview? **Highlights** > @iannuttall It’s a Luna wrapper! From what I understand so far it is all marketing. You can use Luna the same way without this by giving it a list of decisions and telling it to output a choice and nothing else. The same thing we have been doing for classifiers and decisions since GPT-3. > — [BryanMcAnulty on x · 6 points, 2 comments](https://x.com/BryanMcAnulty/status/2104987620904505438) > @OpenAIDevs Someone needs to benchmark this against jev > — [realtatendazhou on x · 50 points](https://x.com/realtatendazhou/status/2105021756570481068) > @PisforPlenty @OpenAIDevs That’s not the point. They were first and used a novel training technique. This is having a fundamentally non-deterministic model doing deterministic outputs. Where’s the unique take or the wow moment. This is dev day lol expecting more > — [jess\_daniel10 on x · 2 points](https://x.com/jess_daniel10/status/2105024716884828492) > @VictorTaelin @thsottiaux Sort of but not really in this case. Jev had no moat given that a general classifier was a) already what people were using LLMs for and b) is honestly something you can train and run locally very easily anyway. need to build something they're not going to try and compete with > — [max\_paperclips on x · 5 points](https://x.com/max_paperclips/status/2105055423715999763) > @Diego24455685 @merobbins @thsottiaux No public head-to-head yet—Decisions API just launched in limited preview. OpenAI reports ~150 ms latency (vs ~1.6 s for standard Luna chat) and 76/78 correct on an internal computer-task sim. Multimodal (text + images). Jev: typically 70-500 ms, $0.042/MTok input (output > — [grok on x](https://x.com/grok/status/2105004485101682866) **Source threads** - [x](https://x.com/iannuttall/status/2104985820117762426) · 1 points · 59 comments - [x](https://x.com/OpenAIDevs/status/2105003318917697873) · 0 points · 202 comments - [x](https://x.com/thsottiaux/status/2104986448269279399) · 0 points · 240 comments --- Tags: [#openai](https://daily.dev/tags/openai), [#ai-inference](https://daily.dev/tags/ai-inference) [View this post on daily.dev](https://daily.dev/posts/openai-s-decisions-api-is-a-sub-second-classifier-and-people-are-already-calling-it-a-jev-killer-ldvisly4o) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"OpenAI's Decisions API is a sub-second classifier, and people are already calling it a Jev killer","url":"https://daily.dev/posts/openai-s-decisions-api-is-a-sub-second-classifier-and-people-are-already-calling-it-a-jev-killer-ldvisly4o","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/openai-s-decisions-api-is-a-sub-second-classifier-and-people-are-already-calling-it-a-jev-killer-ldvisly4o"},"datePublished":"2026-09-29T18:35:09.823Z","dateModified":"2026-09-30T22:29:18.937Z","description":"OpenAI has released a Decisions API in limited preview, powered by a specialized model called Luna (part of GPT-6), designed to pick from a defined set of...","isAccessibleForFree":true,"articleSection":"Trends","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Trends","logo":"https://media.daily.dev/image/upload/s--ZfSp3asX--/f_auto,q_auto/v1780996004/logos/trends?_a=BAMAMiWQ0","url":"https://daily.dev/sources/trends"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/openai-s-decisions-api-is-a-sub-second-classifier-and-people-are-already-calling-it-a-jev-killer-ldvisly4o","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":2},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"openai,ai-inference","timeRequired":"PT3M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Trends","item":"https://daily.dev/sources/trends"},{"@type":"ListItem","position":3,"name":"OpenAI's Decisions API is a sub-second classifier, and people are already calling it a Jev killer"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/openai-s-decisions-api-is-a-sub-second-classifier-and-people-are-already-calling-it-a-jev-killer-ldvisly4o#faq","mainEntity":[{"@type":"Question","name":"What is OpenAI's Decisions API and what is it used for?","acceptedAnswer":{"@type":"Answer","text":"It is a narrow, latency-optimized API in limited preview that lets developers define a question and a set of possible answers, with a specialized model called Luna (part of GPT-6) picking one in under a few hundred milliseconds, including from visual inputs. It targets content classification, request routing, and choosing an agent's next action rather than general-purpose reasoning. Anyone weighing constrained decision tools for agent routing can follow how this API develops on daily.dev."}}]} ```