TypeSafe AI, a Silicon Valley startup founded by former OpenAI researcher Diogo Almeida, has launched Jev, a "decision model" that aims to be a more efficient and cost-effective alternative to large language models (LLMs) like those from OpenAI and Anthropic for certain tasks. Jev, which does not generate sentences or images but instead provides fast, cheap decisions within software applications, has quickly garnered significant attention in the tech industry. It focuses on classification tasks such as approving/blocking requests, routing support tickets, or assessing insurance/credit risk.
The enthusiasm for Jev reflects growing corporate concerns over the cost of deploying AI. TypeSafe bets that many routine tasks currently handled by expensive general-purpose models can be performed more efficiently by specialized systems. Jev is reportedly orders of magnitude cheaper and faster, processing queries for about $0.042 per million tokens, compared to several dollars per million for LLMs. This drastically lower computing cost contributes to the company's profitability.
TypeSafe AI, which concluded a $40 million seed funding round at a $200 million valuation over a year ago, is now fielding investment offers that could value it at $10 billion or more. The company claims Jev is about 100 times cheaper and faster than LLMs, with response times between 70 to 500 milliseconds, compared to 3 to 329 seconds for LLMs. Despite its specialized nature, Jev achieved similar performance to GPT-5.6 Terra and Claude Sonnet 5 on internal classification datasets, while costing significantly less, around $0.0007 per example compared to $0.06 for Terra and $0.12 for Sonnet.
Industry experts acknowledge Jev's potential to address a "latent demand" for models that provide simple, cheap, and fast decisions, an area previously underinvested in due to the focus on higher intelligence in frontier AI models. However, some have questioned how revolutionary Jev truly is, given TypeSafe's secrecy regarding its training methods and the existence of similar classification tools from companies like Meta, Google, and Hugging Face. Nevertheless, platforms like Vercel and OpenRouter have reported immense interest and usage of Jev since its launch.
