OpenAI expands decision model access as competition intensifies in low-cost AI APIs

Traders Union
All News
Business & Wealth
Companies
OpenAI expands Decisions API access

Highlights

OpenAI has launched its Decisions API globally, reducing model response times from up to three seconds to 100–180 milliseconds for speed-sensitive tasks.
Early adopters like Box, Datadog, Ramp, and Speak report major speed improvements, with Speak cutting action times from 1.4 seconds to 0.13 seconds using the API.
Despite OpenAI’s progress, benchmarks such as the AIM System One Index and Artificial Analysis Decision Index currently rate Liquid’s D1 Ultra and Base Lab’s Elephant-1 as higher-performing decision models.

Global rollout targets speed-sensitive AI workloads

As first reported by TechCrunch, OpenAI has made its Decisions API broadly available after first announcing the product at its developer day on September 29. The company says the model is designed for tasks where developers need rapid choices among defined options, with response times reduced from roughly one to three seconds to about 100 to 180 milliseconds.

OpenAI says it adapted its ChatGPT Luna model rather than building a separate calibrated decision model from scratch. Nikunj Handa, product lead for OpenAI’s APIs, said the company also added voice and visual modalities to the public release and adjusted inference resources geographically to reduce latency across regions.

Early customers including Box, Datadog, Ramp and Speak are using the API to replace larger language model workflows where throughput matters more than broader generative capabilities. Speak CTO Andrew Hsu said the product cut the time between actions from 1.4 seconds to 0.13 seconds, while OpenAI says it is also using the technology to improve responsiveness in its Dots voice assistant.

Benchmark pressure grows across the decision model market

OpenAI is entering a segment that has gained traction after TypeSafe’s Jev helped define the category of so-called System One or super-classifier models. These systems are designed to choose between options quickly and return confidence scores, making them attractive for developers seeking lower operating costs and faster execution than standard LLM-based setups.

The competitive picture remains unsettled, with companies weighing intelligence, calibration, speed and cost differently depending on the workload. Public benchmarks cited in the market, including the AIM System One Index and the Artificial Analysis Decision Index, currently favor rivals such as Liquid’s D1 Ultra and Base Lab’s Elephant-1, while the latest Jev model remains competitive and OpenAI’s Decision Model ranks lower on those measures.

OpenAI may hold an advantage in multimodal use cases, particularly image classification, where Handa said Luna shows stronger calibration. He also pointed to further speed gains as a path to new applications, including video-related use cases that could require model calls on every frame.

In our earlier article on KPMG’s CEO survey on AI agents, we noted that large U.S. companies are seeing early efficiency and growth gains, but only a small share of leaders view agentic AI as truly transformational so far. The survey also highlighted that realizing bigger benefits requires continued investment in AI infrastructure, hiring and retraining, and reworking business processes rather than treating AI as a simple plug-and-play upgrade.

This material may contain third-party opinions, none of the data and information on this webpage constitutes investment advice according to our Disclaimer. While we adhere to strict Editorial Integrity, this post may contain references to products from our partners.