Jev decision model arrives on OpenRouter and Vercel AI Gateway
The 'System One' model skips text — typed JSON verdicts with confidence, reportedly ~400x cheaper than an LLM call, callable through gateways you already use. Plus a Next.js patch.
Copy markdown
Answers, not prose
Jev takes a data 'state' plus your questions and returns yes/no or multiple-choice verdicts with confidence scores as typed JSON, evaluating questions in parallel inside a 64K context — no memory, no knowledge base. TypeSafe reports up to ~194x faster and ~445x cheaper than frontier LLMs like GPT-6 Astra on decision tasks; independent write-ups frame it as a 'System One' model built for classification, not generation. For you: intent routing, citation checks, and refund/triage sorting stop costing LLM prices.
Callable through gateways you already use
Jev (v1.13) is now listed on OpenRouter and Vercel AI Gateway, with input reported around $0.042 per 1M tokens on one provider — so you can route to it without onboarding a new vendor. LangChain ships a langchain-typesafe package: pass state and questions to TypeSafeClassifier.invoke() for model routing or pre-tool risk checks at fractions of a cent each. Weekend build: swap one LLM-based classifier in an existing agent for a Jev call and measure the latency and cost drop.
Elsewhere: Next.js patches nine flaws Sept 30
Next.js has pre-announced a scheduled security release for September 30 fixing nine vulnerabilities — one critical, two high, five medium, one low — shipping as 16.3.7 and 15.5.27. Full advisories with affected versions arrive with the release; plan your upgrade window now.