Typesafe AI’s Jev model is being used by cryptocurrency traders and developers to make rapid, narrowly defined decisions, returning probability scores in as little as 70 milliseconds rather than holding a conversation like ChatGPT or Claude.
The system, also known as System One, launched last week after two years of development in stealth mode. Typesafe AI has raised $40 million, and on Sunday it announced that Jev was available to everyone without a waitlist. New users were also given $5 in credits.
Jev is designed to receive information and a tightly framed question before selecting an answer from a list provided by the user. Its responses can be limited to choices such as buy or sell, yes or no, or a numerical score, together with the model’s confidence in that decision.
Typesafe says responses take between 70 and 500 milliseconds. The service costs $0.042 per million input tokens, while output is free.
Unlike conventional chatbots, Jev does not generate prose. That allows it to avoid hallucinating information and makes it easier for other software to use its results consistently. The model has already been applied to real-time video game logic, predictions, parallel information extraction, live scoring and validation.
On Sunday evening, Bitcoin.com News tested Jev using live data from Polymarket’s event, “What price will bitcoin hit in 2026?” The model was asked whether a Binance one-minute candle would reach $85,000, $90,000 or $100,000 before 2027.
At the time, the market had recorded $80,000, while $75,000 and $65,000 had also been reached. Trading volume on the event was close to $67.8 million.
Polymarket users were pricing an 81% chance of $85,000 being reached next. Jev assessed the probability at 55%, followed by 48% for $90,000 and 34% for $100,000. It gave a 50% probability to a fall to $70,000.
The experiment represents only one potential use of the model. Developers in the crypto sector are also testing it for stablecoin strategies, meme coin launch assessments and systems focused on individual tokens.
Jev’s speed has attracted builders looking to place it inside automated trading processes. On 16 September, Jarrod Watts, lead AI engineer at Monad, connected the model to a price feed and asked it to choose between buying and selling. His bot then placed a post-only limit order on Kuru’s onchain order book roughly every 300 milliseconds.
“Jev decides if it should ‘buy’ or ‘sell,’ given the price feed of an asset pair, and executes real trades,” Watts wrote on X.
Watts also released the bot’s code publicly under `jarrodwatts/jev-trader`. Other repositories, forks and ports are emerging across the crypto ecosystem as developers experiment with similar applications.
Jev’s launch points to a possible shift towards AI systems that provide fast, bounded judgements rather than conversation. Its usefulness will depend on how far markets and software developers are prepared to rely on a confidence score.
