Hacker News front pageSep 16, 2026, 03:25 UTC+8

TypeSafe launches Jev, a structured-decision model that’s 40–400× cheaper and 20–200× faster than frontier LLMs

Jev: New frontier model 40-400x cheaper and 20-200x faster

TypeSafe founder Diogo Almeida (ex-OpenAI) announced System One models and the first public model Jev. Jev doesn’t generate strings—it outputs type-safe structured values with calibrated probabilities, making hallucinations and type errors mathematically impossible. Input costs $0.042/MTok, output is free; end-to-end latency is 70–500ms, 40–200× faster than GPT-5.6 Terra. The training method, RLCD, optimizes for calibrated decisions rather than human preference. A side-by-side demo with GPT-5.6 Terra shows only one disagreement—on churn likelihood—which the author says is genuinely ambiguous. I’d hold off on full enthusiasm: the post doesn’t provide independent third-party benchmarks, and long-term pricing sustainability isn’t proven yet.

Read the original ↗Export Markdown

Take

The reason to click is the claim: Jev makes hallucinations and type errors mathematically impossible by never generating strings—only typed structured values with calibrated probabilities. Diogo Almeida, ex-OpenAI, says input costs $0.042 per million tokens, output is free, and end-to-end latency is 70–500ms, which he pegs at 40–200× faster than GPT-5.6 Terra. The training method, RLCD, optimizes for calibrated decisions rather than human preference.

I'd discount this a bit for now. There are no independent third-party benchmarks—every number comes from TypeSafe's own blog post. The speed and cost comparisons are against GPT-5.6 Terra, but the post doesn't specify the task, input length, or throughput methodology. "Output is free" sounds like "too cheap to meter," but long-term sustainability depends on inference cost structure, which isn't disclosed.

The more interesting bit is the positioning: Jev isn't a chatbot or code generator. It's a decision component that takes unstructured state in and spits out typed probabilistic values—classify, route, score, extract. If the accuracy really matches frontier LLMs on System One tasks, it could be compelling for low-latency, high-reliability, high-volume pipelines. But right now it's one blog post. I'd wait until someone runs it on a real workload.

Source: Hacker News front page · typesafe.ai
Published: Sep 16, 2026, 03:25 UTC+8 · 14 days ago