TypeSafe has launched the Jev model, sparking a new category of decision-making AI. The release has drawn major industry players, including Microsoft with its Microsoft Decision, OpenAI with the Decisions API, and Fastino with the Glide model. Additionally, Cloudflare and Perplexity have introduced open-weight solutions that allow users to run models on their own hardware.
Unlike standard large language models, these decision-making tools do not generate text token by token or maintain conversation history. Instead, they return probability results and support parallel processing. TypeSafe claims that Jev operates 193 times faster than standard LLMs and reduces costs by a factor of 445.
The launch of Jev has triggered allegations of plagiarism. Nandakishor Mukkunnoth, founder of ConvAI Innovations, states that his SalesRLAgent paper, published on March 30, 2025, already introduced a probability estimation model with publicly available weights and datasets.
Mukkunnoth noted that his previous work focused on specific sales scenarios, which limited its visibility compared to TypeSafe's general-purpose engine. TypeSafe offers Jev as a fully managed product, complete with API access, documentation, developer kits, and a pricing structure.
In response to the controversy, Mukkunnoth released an open-source model named Laya on the same day Jev was launched. Laya is based on the BERT encoder-only architecture, supports multiple languages, and is compatible with Jev-style interfaces for local deployment.
