Unlike large language models (LLMs) that generate text, the competitive landscape for decision AI models focused on probability-based judgment has been reshaped into a multi-party race in a short period of time.

After TypesafeAI, founded by former OpenAI employee Diogo Almeida, unveiled Jev in September, major companies such as OpenAI, Amazon Web Services (AWS) and Cloudflare also jumped into a race for decision models similar to Jev.

Given a complex situation, narrowly scoped questions and a list of answer options, Jev provides decisions and probabilities rather than sentences. The company says it will be an AI model suited to AI agents.

TypesafeAI says what AI agents need are decisions such as whether something is urgent, whether to continue or stop, and whether to allow or block. With an LLM, it first writes a sentence and then converts that sentence into a decision. Many of the repeated tasks inside an agent do not require writing sentences. That is why TypesafeAI positions Jev as an alternative that overcomes the limitations of LLMs.

According to the related industry, Jev has drawn significant attention in the AI community since its release. The fact that leading tech companies are introducing models based on a structure similar to Jev appears to be related.

ㆍTypesafeAI, the AI startup in the spotlight, seeks to raise $1 billion in funding... valuation $10 billion?

OpenAI in late September unveiled a limited release of its "Decisions API" at its annual developer event, designed to make quick judgments within a fixed set of options. It is a product similar to Jev.

ㆍOpenAI unveils 'Decisions API' to support fast judgment… targeting software automation

OpenAI CEO Sam Altman explained that the Decisions API works by giving OpenAI's "Luna" model preset options and having it choose within them. It can select image classification labels or choose one among multiple agent actions. "If you narrow the range of choices, you can achieve very fast speed while still maintaining image understanding, broad language support, and safety guardrails," he said.

AWS also released an open-source lightweight decision model called "Strands Decider 2B" aimed at developing agentic AI. TypesafeAI did not disclose Jev's structure.

ㆍAWS also launches open-source decision model 'Strands Decider 2B'

Strands Decider 2B also does not generate text and operates by choosing one option from a predefined set, and was designed to reduce token consumption and response latency. It is also a concept similar to Jev.

Cloudflare also introduced a decision AI model called "Clef," claiming it is smarter and faster than Jev.

According to a report by The Register, Clef is offered in two versions: the base model Clef and the smaller, faster Clef Flash. Like Jev, the two models can answer three fixed-answer question types, including yes-no, multiple choice and ranking.

Cloudflare developed Clef by further training the open-source LLM Qwen for decision-making. It used Qwen3.8-27B for Clef and Qwen3.5-9B for Clef Flash.

Clef classifies images and video as well as text. It can handle up to 64,000 tokens.

Clef can be used on Cloudflare's AI model inference platform Workers AI. It costs $0.24 per 1 million tokens, nearly six times more expensive than Jev ($0.042). If the cost is burdensome, users can download the model from Hugging Face and run it themselves. The licence is Apache 2.0, the same as Qwen. Michelle Chen (미셸 첸), product manager for Cloudflare's AI Platform Group, said the training dataset was not disclosed.