Ollaya runs decision models locally, up to 255 options

Ollaya follows the release of Jev, a decision model that appeared on September 15, 2026, and drew wide attention for generating judgments and confidence scores across multiple options quickly and at low cost. The article cites its speed and low cost relative to frontier models like Claude and GPT series, and says applications built on Jev have been developing quickly, with new decision models inspired by its ideas appearing one after another. Many of these decision models are open, meaning they can be run locally.

Ollaya is modeled on what Ollama did for LLMs: making local execution simple. It runs on Windows, macOS and Linux, and the article notes that Ollama's version 0.35 added decision-model support but was limited to three models at the time of writing, while Ollaya can run more capable ones. The option count also differs, with Ollama capped at 15 per request and Ollaya at 255.

The article's comparison suggests local execution can approach Jev's accuracy and shorten latency for some models, though not uniformly. The practical test for readers weighing local decision models is likely their own GPU and workload, since the example used a GeForce RTX 4070 and latency figures depend on the model chosen. Whether local setups narrow the gap further is something to watch as more models appear.