Technology 6 min read By Alice Ashford
TypeSafe's Jev AI model goes viral in Silicon Valley as OpenAI races to respond
TypeSafe AI's Jev model, built for fast classification decisions, has been adopted rapidly by developers. OpenAI launched a competing product within three weeks, signalling a new front in the race to automate routine business judgments.
TypeSafe AI's Jev, a model built specifically for rapid classification decisions, has become a viral hit among Silicon Valley developers since its September release. Within a day of appearing on Vercel's AI Gateway, nearly 13 per cent of the platform's paid teams had experimented with it, and Vercel said Jev reached more than twice as many paid teams in its first 24 hours as any previous model launch.
The enthusiasm reflects a practical gap in the artificial intelligence market. Businesses increasingly want to automate thousands of routine judgments each day — sorting customer messages, categorising documents, flagging errors — and at that volume, small differences in cost, speed and accuracy determine whether automation is worthwhile. Jev is designed to handle those judgments faster and more cheaply than general-purpose language models while retaining the flexibility of natural-language instructions.
OpenAI responded swiftly. On 6 October, just three weeks after Jev's launch, it rolled out a competing product called Decisions API, built on its existing GPT-6 Luna model. The move underscores how quickly the market for practical, high-volume AI decision-making is becoming a battleground between specialised startups and the industry's largest players.
TypeSafe chief executive Diogo Almeida began thinking about Jev after helping develop the technology behind ChatGPT at OpenAI. He saw a gap between models' ability to answer questions and their usefulness in automating routine work, and said he wanted to address AI's «massive over-promise under-deliver issue» and avert an «AI winter». Building a company, he said, «just happened to be the most effective way to do that».
Almeida's reasoning was that in a future where AI had transformed the economy, the vast majority of requests to models would come from code running automatically rather than from people. A customer-service program, for example, could ask a model whether an incoming message concerns a billing issue or a technical problem, then route it to the appropriate team. Developers could set a confidence threshold: above it, the message is routed automatically; below it, a human checks.
Such classification tasks are not new. Rudimentary machine learning systems have handled them for decades, and email services have long used simple models to filter spam. But those systems tend to be inflexible, lack deep understanding of text and require technical know-how to build. Large language models improved on them by understanding text more deeply and allowing developers to change instructions for different jobs, but they consume more computing power, making them slower and more expensive at scale.
TypeSafe says Jev combines the speed and cost advantages of simple classifiers with the natural-language setup of large language models. Developers describe what they want evaluated, and Jev returns choices, scores or probabilities their software can use. The company says it produces those outputs together, avoiding the time spent generating an answer piece by piece. Developers must still establish whether its judgments are accurate enough for their particular task.
Almeida initially thought making the models behind Jev reliable would take a week. Instead, TypeSafe spent two years developing it. «Making the models reliable was so much harder than expected,» he said.
Early users have been impressed by its speed. Niels Mouthaan, the independent developer behind the Daily time-tracking app, tested Jev in a spelling and grammar prototype that gave users feedback as they typed. He said users expect «close-to-instant feedback while typing» and was impressed by Jev's speed, seeing promise in using it to flag errors while a language model suggests corrections only where needed.
TypeSafe calls Jev a «System One» model, borrowing the term for fast, intuitive judgments from psychologist Daniel Kahneman. Ashutosh Mathore, a founder of environmental compliance software startup Atlensa, also experimented with Jev and praised its speed.
The competitive dynamics are already shifting. OpenAI's Decisions API uses an existing model rather than one trained specifically for classification, a choice that may trade some efficiency for faster deployment. TypeSafe's bet is that dedicated training produces meaningfully better results for the narrow but vast market of automated routine decisions.
For British businesses weighing automation, the contest matters. The cost and latency of each AI call accumulate quickly when a company processes thousands of customer messages or documents daily. A model that is faster and cheaper per decision could widen the range of tasks worth automating, from routing support tickets to categorising internal records. The arrival of a major competitor within weeks of Jev's launch suggests both companies expect that market to grow rapidly.
4
