TypeSafe introduces Jev, claiming up to 200x faster decisions than frontier LLMs

TypeSafe, an AI startup focused on software automation, is introducing a new model called "System One," which it said is designed to make fast, structured decisions that software can use directly.

The company on Tuesday launched Jev, its first publicly available model built on System One, in early access.

In a blog post, Diogo Almeida, founder and CEO of TypeSafe, said that the team had spent its time in stealth building a new AI stack focused entirely on automation. The stack includes a new model architecture, a parallel sampler for maximum efficiency, and a training method the company calls Reinforcement Learning for Calibrated Decisions (RLCD).

TypeSafe said current frontier models can hallucinate and introduce unnecessary variability into systems that depend on predictable behavior. The startup said it is taking a different approach by building intelligence that can "serve as a composable software primitive for applications requiring semantic judgment and intelligent decision-making."

"Think of Jev as a frontier-intelligence function call: unstructured state in, typed probabilistic decisions out," said Almeida, who helped develop OpenAI's ChatGPT.

Faster without generating text

Unlike traditional large language models, Jev does not generate text or strings, and is instead optimized for structured outputs. The founder said Jev can achieve similar levels of intelligence to existing large language models while "being two orders of magnitude faster and more efficient."

TypeSafe estimated Jev's end-to-end response time at around 70ms to 500ms, compared with roughly 3 to 329 seconds for existing frontier LLMs.

"This can range from 40 times to 200 times faster for the same levels of frontier intelligence for System One shaped queries," the founder wrote in the post. Existing frontier models are "fast enough for interfacing with humans, but a big bottleneck when integrated in code," he added.

"We started TypeSafe because we believe that AI needs an interface software could depend on," Almeida said.

TypeSafe received $40 million in seed funding led by DCVC, according to the statement.

"TypeSafe is approaching one of the biggest remaining challenges in AI: turning increasingly capable models into technology that developers can reliably build into products at scale," James Hardiman, general partner at DCVC, said in the statement.