Decision models are suddenly everywhere. OpenAI’s is now public.
The New Stack Meredith Shubel ● Covered by 3 sources
OpenAI just opened its decision model to all developers and added image input. The twist: rivals are shipping open weights, while OpenAI is betting on a hosted API.
Based on reporting by The New Stack, Meredith Shubel — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
OpenAI has put its Decisions API into public beta, and the timing is no accident. The company is billing the model at $0.10 per 1M input tokens, with no charges for output tokens or cache reads and writes. It runs on GPT-6 Luna, and it can turn text or images into three kinds of structured output: predicates, choices, and scores.
That image support is the part OpenAI wants people to notice. In the company’s own demo, the API looks at frames from a road scene and decides whether a car should change lanes or keep going. OpenAI says the same model can make that call in a fraction of the time a reasoning model would need, which is the kind of pitch that lands well when speed matters more than philosophical satisfaction.
The company also showed the API working with Hugging Face’s upcoming programmable Microduck robot, an open-source bipedal developer robot. Paired with GPT-Live, OpenAI’s real-time voice model, the decision model watches camera frames and chooses where the robot should look. In the example shown, “follow the apple” becomes a chain of steps: identify the fruit, then turn toward it. OpenAI says the same setup can handle looser prompts too, like “follow the fruit.”
And then there’s the other half of the story: everyone else seems to be heading the opposite way. Perplexity, Cloudflare, and Amazon all launched their own decision models this month, but those are open-weight releases that developers can download and run themselves. Perplexity’s pplx-decider-v1-27b, fine-tuned from Qwen3.8-27B and released under Apache 2.0, claims a better overall score than TypeSafe’s Jev across 11 benchmarks, though Jev wins six of them.
Amazon’s Strands Decider 2B is built on Qwen3.5-2B and is aimed at roughly 2 billion-parameter models. Cloudflare has gone even more open with Clef and Clef-flash, both Apache 2.0 weights and also available through Workers AI. OpenAI’s bet is different: keep the model hosted, add vision, and hope developers value convenience enough to pay for it instead of running their own.
My take — AI-written commentary, not fact-checked reporting
The real split here isn’t intelligence, it’s control. OpenAI is selling convenience; everyone else is handing developers the keys and letting them park in the driveway. For a category called decision models, the biggest decision may end up being whether anyone wants another hosted gatekeeper.
Read more about this at: The New Stack