jev-latest — API, Pricing & Context Window | Vivgrid

jev-latest is TypeSafe's System One model, and it does not write text. You send it the current state plus the questions you need answered, and it returns typed answers — a picked option, a score on your rubric, or a yes/no probability — each with a calibrated confidence value.

That makes it the right tool for the decisions an agent makes around the model call: routing a ticket, grading a retrieval hit, deciding whether a step is done, gating a retry. Those are classification problems, and running them through a frontier chat model means paying for generated tokens and parsing prose back into a type. Jev answers in 70–500 ms, typically 40–200× faster.

Specifications

Pricing

Pricing in USD per 1M tokens.

Quick start

Call jev-latest through Vivgrid's unified /systemone endpoint. Get an API key from the Vivgrid Console.

curl https://api.vivgrid.com/v1/systemone \
  -H "Authorization: Bearer $VIVGRID_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "jev-latest",
    "state": "The website has been down for 2 hours and customers cannot complete purchases.",
    "questions": {
      "routing": {
        "type": "choice",
        "instructions": "Which team should handle this?",
        "criteria": {
          "billing": "Payments, invoicing, refunds",
          "technical": "Bugs, outages, integrations",
          "sales": "Pricing, upgrades, new accounts"
        }
      },
      "severity": {
        "type": "score",
        "instructions": "How severe is this issue?",
        "criteria": ["Minor", "Moderate", "Critical"]
      }
    }
  }'

Question types

• Choice — pick one option from a map of option → rubric description
• Score — rate the state against an ordered array of levels, returning a continuous score plus its distribution
• Noul — a yes/no question, returning the probability the statement is true

A single request may carry many questions; state plus the longest question must fit in 32K tokens, and the whole request in 64K.

Ideal use cases

• Routing, triage, and intent classification in agent pipelines
• Guardrails and gating decisions between tool calls
• Scoring retrieval results or model output against a rubric
• High-volume labeling where per-call latency and cost dominate

Related models

• gemini-3.8-flash — a fast general-purpose model when you need generated text
• glm-5.3-flash — low-cost chat completions for lightweight classification