The OpenAI Decisions API is now open to every developer, and the pitch is simple: pick the best option fast, and say how sure you are. OpenAI says response times drop from one to three seconds to roughly 100 to 180 milliseconds. It became generally available on October 22, 2026, according to TechCrunch.
Here are the facts you need before the details:
- Generally available since October 22, 2026 (first announced at OpenAI’s developer day on September 29).
- It chooses between options you supply and returns a confidence rating, instead of writing free text.
- OpenAI’s claimed speed is about 100 to 180 ms per response.
- Public version adds voice and visual input.
- Pricing was not published in the launch coverage.
The big takeaway: this is a speed tool for narrow jobs, not a smarter chatbot.
What the OpenAI Decisions API actually does
A normal language model writes text. A decision model does something narrower: you hand it a list of options, and it picks one and tells you how confident it is. Think of it as a very fast classifier.
The category was created by TypeSafe with its Jev models. OpenAI is now the biggest name to join in. For example, an app could use it to route a support ticket, choose the next step in a voice chat, or tag an image in real time.
OpenAI didn’t train a brand-new model for this. Per TechCrunch, it reconfigured its existing ChatGPT Luna model to run inference faster and cheaper.
Why 100 milliseconds changes what you can build
Speed is the whole story here. At one to three seconds, an AI call feels like waiting. At around a tenth of a second, it feels instant, and you can put it in the middle of an app without users noticing.
Speak, which builds AI language tutors, cut the gap between actions from 1.4 seconds to 0.13 seconds. Box, Datadog and Ramp also tested the API, swapping it in for OpenAI’s regular LLMs to get higher throughput.
OpenAI product lead Nikunj Handa also mentioned a goal of about 60 milliseconds. That would be fast enough for one API call per video frame, though it isn’t shipping yet.
Speed has been a theme this month. We looked at the same race from another angle in our post on Mercury 3, a diffusion LLM built for 1,000+ tokens a second.
How it stacks up against Jev and other decision models
OpenAI isn’t winning the benchmarks. Here is where TechCrunch says the models rank on two decision-model indexes.
| Model | AIM System One Index | Artificial Analysis Decision Index |
|---|---|---|
| OpenAI Decision Model | 8th | 14th |
| TypeSafe Jev (latest) | 7th | 3rd |
Liquid’s D1 Ultra and Base Lab’s Elephant-1 rank highest on both. TypeSafe’s models are also the most-queried decision models on OpenRouter, and Jev ranks first on JevMan, a benchmark where decision models play PacMan.
So OpenAI’s advantage isn’t raw rank. It’s distribution: millions of developers already have an OpenAI account, and many will try this first. TypeSafe, valued at $7.6 billion, released Jev 1.14 the week before with image support.
Where a decision model beats a regular chatbot call
The two tools do different jobs, and mixing them up wastes money. This comparison shows the split, using OpenAI’s own speed numbers.
| Feature | Regular LLM call | Decision model |
|---|---|---|
| Output | Free-form text | One chosen option plus confidence |
| Typical response time | 1 to 3 seconds | About 100 to 180 ms (OpenAI’s claim) |
| Best for | Writing, reasoning, chat | Routing, labeling, quick choices |
If your app only needs a yes, no or pick-one answer, a regular model is overkill. The confidence score also lets you hand low-certainty cases to a bigger model or a person.
What we still don’t know
The launch coverage leaves a few practical gaps. These matter before you move anything into production:
- Pricing: described as cheap, but no figures were given.
- Rate limits and usage caps were not specified.
- Image outputs: customers want more flexibility, and OpenAI hasn’t added it yet.
- Quality: results depend on your workload, and TypeSafe’s CEO urges testing models on your own tasks.
That last point is the safest advice. Run a small test with your real data and compare cost and accuracy before you switch. OpenAI has also been tweaking API costs elsewhere, as in its 12-hour prompt cache pricing, so watch the pricing page.
Frequently Asked Questions
What is the OpenAI Decisions API?
It’s an API that chooses between options you provide and returns a confidence rating. It’s built for fast, narrow choices rather than long text answers.
When did the OpenAI Decisions API launch?
OpenAI announced it on September 29, 2026, and opened it to all customers on October 22, 2026.
How much does it cost?
OpenAI hasn’t published figures in the launch coverage. It is described as cheap, so check OpenAI’s pricing page before you plan a budget.
Is it a new AI model?
Not exactly. OpenAI reconfigured its existing ChatGPT Luna model to run faster and cheaper for decision tasks.
Should I replace my normal LLM calls with it?
Only for simple picks like routing or labeling. For writing, reasoning or anything open-ended, stick with a regular model.
If your app makes lots of small choices and users feel the lag, the OpenAI Decisions API is worth a weekend test. Skip it if your calls are rare or need real reasoning, because a faster classifier won’t fix a slow idea. And with Jev and others ranking higher, benchmark at least two options before you commit.
Leave a Reply