TypeSafe AI launches Jev, claiming 20-200x speed boost over LLMs by skipping token generation
TypeSafe AI, founded by former OpenAI researcher Diogo Almeida, launched Jev, a new AI model that skips token-by-token generation to make structured decisions. Jev returns typed answers with confidence scores in 70-500ms, claiming 20-200x faster speeds and 40-400x cost savings over traditional LLMs. Priced at $42 per billion input tokens with free outputs, Jev is available in beta on OpenRouter and targets software workflows like classifying support tickets.
Editorial responsibility
- No named human review is recorded for this page.
- Reports are grouped by semantic similarity and deterministic rules. Language models may assist titles, summaries, translation and cross-source analysis; the page reads the event directly, while its address stays stable when the title changes.
- Summary covers the current reports
Cross-source coverage
Reporting timeline
Jev AI model by TypesafeAI launches in beta on OpenRouter platform
Jev, an AI model developed by TypesafeAI, is now available in beta on the OpenRouter platform. Unlike traditional text-generating models, Jev is described as a 'System One' model that takes an application's state and a typed question as input, and returns a typed decision with an associated probability. The announcement highlights that no JSON prompting, parsing layer, or validation is required to use the model. This release marks a new type of AI inference tool for developers, focusing on structured decision-making rather than open-ended text generation.
Jev AI model built for decisions claims 20-200x speed and 40-400x cost savings over LLMs
A new AI model called Jev, introduced by Diogo Almeida and trained using a novel method named RLCD, is claimed to be 20 to 200 times faster and 40 to 400 times cheaper than comparable large language models (LLMs). According to the announcement, Jev costs $0.042 per million input tokens with free outputs, but it cannot generate text. The model is designed specifically for decision-making rather than text generation, with demos including controlling a game and choosing Wikipedia links. The stated goal is to embed fast, cheap AI decisions into software. The claims are attributed to the source and have not been independently verified.
Read sourceTypeSafe AI launches Jev, a fast and low-cost System One decision model
TypeSafe AI has introduced its first model, named Jev, described as a small and customizable System One decision model. Unlike large language models (LLMs), Jev cannot reason or write explanations; instead, it generates rapid judgments with calibrated probabilities and a confidence score. Complex tasks are broken down into separate questions and recombined in code. TypeSafe reports a response time of approximately 150 milliseconds per call, at around 100 times lower cost and faster responses than LLMs. Pricing is set at $42 per billion input tokens, with free output.
Show 2 older updatesHide older updates
Diogo Almeida's Typesafe AI releases Jev, a model that refuses to chat for faster decisions
Diogo Almeida, who helped build Reinforcement Learning from Human Feedback (RLHF) at OpenAI, has launched a new company called Typesafe AI and released a model named Jev. Unlike standard generative AI models, Jev is designed to refuse open-ended chat and instead act as a pure API for structured decisions. It returns typed answers with exact confidence scores, stripping out generative prose entirely. The company claims Jev achieves 150-millisecond response times, making it 200 times faster than standard models. Pricing is set at $42 per billion input tokens, with output tokens being free forever. The post's author opines that this is a smart way to route logic without paying for unnecessary words, highlighting a potential shift in agentic workflows where text generation is seen as a bottleneck.
Read sourceTypeSafe AI launches Jev, claiming 20-200x speed boost over LLMs by skipping token generation
TypeSafe AI has launched Jev, a new AI model that claims to be 20-200 times faster than traditional large language models (LLMs) by completely skipping token-by-token generation. Instead of generating text sequentially, Jev is designed to make software decisions directly: given data and a predefined question, it returns a typed choice or score with probabilities and confidence. The company reports response times of 70-500ms and performance 40-200x faster than comparable LLMs, with output tokens free, making it 40-400x cheaper. Jev evaluates multiple structured questions in parallel, acting as a smart if-statement for fuzzy judgments in software workflows, such as classifying support ticket urgency or policy violations.
Read source