OpenAI officially expanded access to its Decisions API for all users and developers globally on October 22, 2026. The move marks an overt step by the research lab to recapture market share in the low-cost, ultra-fast intelligence model segment, where TypeSafe AI’s Jev model, founded by former OpenAI engineers, has rapidly carved out an entirely new product category.
Latency Optimization from the Existing Luna Model
Since it was first announced at the DevDay conference in late September 2026, the commercial version of the Decisions API has gained the ability to accept voice and image inputs. Instead of training a dedicated decision model from scratch, OpenAI’s engineering team restructured the ChatGPT Luna model to maximize response speeds and slash compute costs for classification tasks.
Nikunj Handa, OpenAI’s Head of API Product, shared that the biggest breakthrough lay in compressing response times from roughly 1 to 3 seconds down to just 100 to 180 milliseconds. The development team’s next goal is to reach the 60-millisecond threshold, unlocking the ability to make continuous API calls for every frame in a live video stream. When the Decisions API was initially introduced, the product primarily served internal safety filtering layers, but demand to replace expensive large language models in agent orchestration pipelines accelerated widespread commercialization.
Real-World Impact Across Enterprise Customers
In early trials with partners such as Box, Datadog, Ramp, and Speak, engineers replaced traditional text generation models with the Decisions API to boost throughput. A representative from language-learning app Speak noted that the switch reduced wait times between AI actions from 1.4 seconds to just 0.13 seconds, creating a nearly instantaneous, natural conversational experience. OpenAI itself also uses this solution to improve response times for its Dots voice assistant.
The surge in decision models reflects Jevons’ paradox in economics: as compute resources become cheaper and more efficient, the volume of practical applications explodes exponentially rather than decreasing. Tasks such as data routing, text classification, or triggering conditional logic in software no longer require incurring the high costs of sequence-generation models with hundreds of billions of parameters.
Fierce Competition on the Leaderboards
Although OpenAI quickly rolled out the API, market leadership in this segment remains fiercely contested. Following TypeSafe’s $870 million funding round, the Jev model has continued to maintain leading query volumes on distribution hubs like OpenRouter. On dedicated benchmarks such as the AIM System One Index and the Artificial Analysis Decision Index, models like Liquid’s D1 Ultra and Base Lab’s Elephant-1 currently hold top spots, while OpenAI’s model ranks eighth and fourteenth, respectively.
Nevertheless, OpenAI’s greatest advantage lies in its multimodal processing capabilities. Thanks to Luna’s robust vision foundation, the tool demonstrates superior accuracy when classifying complex images. Competition between purpose-built decision models and ultra-fast fine-tuned foundation models is expected to continue reshaping the operational infrastructure of the entire AI agent ecosystem moving forward.