Laya the open source version of Jev

Open-source decision models with sub-millisecond latency and multilingual support are now available for self-hosting.

Nandakishor Mukkunnoth, Founder & CEO of ConvAI Innovations, has released Laya, an open-source System 1 decision engine that he claims is a direct competitor and improvement over TypeSafe AI's proprietary Jev model. Mukkunnoth asserts that Jev, launched in September 2026 by a lab co-founded by a ChatGPT co-inventor, introduced concepts previously detailed in his own research and publications from March 2025 and September 2025. Laya is described as a non-autoregressive model that provides lightning-fast, calibrated probability predictions over structured schemas, designed for tasks like routing, classification, and scoring. It offers sub-35ms latency on a single GPU, supports over 100 languages, and is available under an Apache 2.0 license with open-source weights, contrasting with Jev's metered API costs and closed nature. Benchmarks provided by ConvAI show Laya achieving higher accuracy and significantly lower latency compared to Jev across several tasks, including AG News and DAIR Emotion, with a 3x better calibration error.

Laya's release introduces a significant challenge to the burgeoning market for specialized AI decision engines, particularly those developed by well-funded frontier labs. By offering a fully open-source, high-performance alternative to proprietary systems like Jev, ConvAI Innovations is democratizing access to rapid, calibrated decision-making capabilities. This move pressures competitors to either open their own models or risk being outmaneuvered on both cost and accessibility, potentially accelerating adoption of these efficient AI components across a wider range of applications.

The key differentiator for Laya appears to be its robust multilingual routing and the explicit focus on calibrated probabilities, addressing known failure points in large language models for structured decision tasks. The ability to efficiently route queries across over 100 languages without significant overhead, combined with its raw speed and open weights, positions Laya as a compelling option for developers seeking to integrate reliable, fast, and cost-effective AI into their systems. This development could lead to a significant shift in how businesses approach automated decision-making, moving away from generalized LLMs towards specialized, efficient models for specific tasks.