OpenAI Decisions API and Jev: Multimodal Decision-Making and Cost Comparison

Clip title: OpenAI’s Decisions API just dropped. Here’s how it compares to Jev. Author / channel: Mark Kashef URL: https://www.youtube.com/watch?v=uTU5Ihgl_7Q

Summary

This video introduces OpenAI’s new Decisions API, a service designed to evaluate text and images to generate structured decisions for applications. The presenter, Mark Kashef, details how the Decisions API works and then conducts a head-to-head comparison with Jev, a prominent open-source alternative, evaluating them on criteria such as cost, speed, and performance across various practical scenarios.

The Decisions API functions by taking either text or an image, allowing users to pose a question about the provided “evidence.” It then returns an answer that an application can readily use. The API supports three main types of questions: true/false (which OpenAI calls a “predicate,” similar to Jev’s “Noul”), multiple-choice, and “how much” questions where it selects an appropriate level based on a rubric. Crucially, the Decisions API, powered by models like GPT-6 Luna, provides a confidence score (probability) for each answer, indicating the likelihood of its correctness.

In a direct comparison, several key differences and similarities between the Decisions API and Jev emerge. The most significant differentiator is that OpenAI’s Decisions API is multimodal, capable of processing both text and images as evidence, whereas Jev is currently limited to text. Both services offer similar answer types (yes/no, choice, score) and can handle multiple questions per request. However, a major divergence lies in cost: Jev is significantly cheaper at $0.042 p er mi ll i o nin p u t [[co n ce pt s / t o k e n s ∣ t o k e n s]] co m p a re d t o t h eDec i s i o n s A P I^{'} s$ 0.10, making Jev approximately 58% more cost-effective. Another notable difference is that the Decisions API can refuse to answer certain questions, particularly in image understanding tasks, to avoid potential abuse, while Jev consistently provides an answer within the predefined choices.

Practical demonstrations revealed nuanced performance characteristics. For image inspection tasks, the Decisions API efficiently identified visible damage, no damage, or indicated an inability to assess the image, demonstrating its unique multimodal capability. In text-based scenarios like website outage investigations and business report verification, both services achieved 100% accuracy in the tested claims. The Decisions API generally showed slightly faster average API call times (e.g., 146ms vs 155ms for report verification), but Jev often had faster median investigation times and was consistently cheaper.

In conclusion, the choice between OpenAI’s Decisions API and Jev depends on specific application needs. If multimodal input, particularly image analysis, is a requirement, or if slightly faster average response times for individual calls are paramount, the Decisions API is the clear winner. However, for applications that primarily deal with text-based evidence and prioritize cost-efficiency, especially at scale, Jev presents a compelling and significantly cheaper alternative with comparable accuracy and competitive performance. The AI landscape for decision-making continues to evolve rapidly, and these services offer robust tools for developers looking to integrate intelligent classification into their products.

Video Description & Links

Description

Work With Us: https://www.promptadvisers.com/

OpenAI just released the Decisions API, their take on what Jev does. You give it some text or an image, ask it a question, and it sends back an answer your app can use. I walk through how it works, what goes into a request, and how it compares to Jev on capability, cost and speed.

Then I put both through three hands-on demos. First, an image inspection that only the Decisions API can run, since Jev takes text only. Then a website outage investigation and a business report check, where both get the same question, the same evidence and the same choices. You’ll see the response times, the actual request costs and where each one comes out ahead.

The short version. If you want cheaper, Jev. If you want multimodal and faster, the Decisions API.

CHAPTERS 00:00 OpenAI’s Decisions API vs Jev 00:31 What the Decisions API does 00:43 Three ways to ask a question 01:04 The fastest way to try it 01:19 Probabilities, predicates and Noul 01:35 How a request flows 02:03 A product return example 02:55 Decisions API vs Jev side by side 03:44 Cost per million input tokens 04:03 Refusals 04:22 Demo 1, image inspection 05:21 Demo 2, website outage investigation 07:19 Demo 3, report verification 08:10 Results on speed and cost 08:59 Which one should you choose 09:12 Final thoughts

A practical comparison of OpenAI’s Decisions API and Jev by TypeSafe, covering predicate, choice and score questions, GPT-6 Luna, pricing per million input tokens, response times, image input and when to use each AI decision API.

OpenAI DecisionsAPI Jev

URLs

https://www.promptadvisers.com/

OpenAI — Wikipedia
Jev
Mark Kashef
GPT-6 Luna — Wikipedia
Noul — Wikipedia

NemoClaw Knowledge Wiki

Explorer

OpenAI Decisions API and Jev: Multimodal Decision-Making and Cost Comparison