Summary
This video introduces Jev, a new AI model from Typeset (founded by a former ChatGPT co-founder) that returns decisions as probabilities instead of generated text, making it dramatically faster and cheaper than frontier chat models like GPT-6 Astra. The creator runs side-by-side speed and cost tests, showing Jev completing tasks like spam filtering, email triage, and ownership classification in roughly 3 seconds versus around 10 seconds for Astra, at a fraction of the cost across batches of 1,000 requests.
Beyond raw benchmarks, the video walks through practical applications: routing support tickets, scoring buying intent, triaging intake forms, and even building a real-time “AI slop” detector for social feeds. It explains Jev’s three output types — yes/no decisions, multiple-choice selections, and 1-100 scores — and how developers can access it through OpenRouter with an API key.
Viewers also see Jev combined with larger models like GPT-6 Astra and Claude Fable 5.1 in a hybrid workflow, using Jev for fast classification and the larger models for generation. The video closes with a free downloadable skill/prompt pack for setting up this combined pipeline, aimed at builders looking to cut latency and inference costs in AI-powered products and internal tools.
📺 Source: Jack Roberts · Published September 19, 2026
🏷️ Format: Hands On Build
