Jev on OpenRouter: Model typesafe/jev-1.13

OpenRouter added Jev on September 18, 2026, three days after TypeSafe’s launch. Because Jev returns decisions rather than text, OpenRouter serves it through a dedicated Decisions endpoint that is still marked alpha.

At a glance

Item	Value
Model	typesafe/jev-1.13 (provider: TypeSafe); alias ~typesafe/jev-latest follows the newest Jev
Endpoints	POST https://openrouter.ai/api/alpha/decisions (OpenRouter’s format, alpha); POST https://openrouter.ai/api/v1/systemone (TypeSafe’s format)
Auth	Authorization: Bearer <OpenRouter key>
Price	$0.042 per 1M input tokens, $0 per 1M output tokens
Context window	32,000 tokens
Free tier	None for Jev; you need credits on the account
Question types	noul, choice, score (same names as TypeSafe)

Why Jev is “missing” from the models list

If you query GET /api/v1/models, Jev does not appear. By default that list covers models that produce text; Jev’s output type is decisions. Add a filter to see it: GET /api/v1/models?output_modalities=decisions returns both typesafe/jev-1.13 and ~typesafe/jev-latest. Do not send Jev requests to /api/v1/chat/completions.

Request format

The body is almost the same as on TypeSafe’s own API: model, state and a questions map. This example follows OpenRouter’s API reference:

curl -s https://openrouter.ai/api/alpha/decisions \
  -H "Authorization: Bearer $OPENROUTER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "typesafe/jev-1.13",
    "state": {
      "message": "My invoice lists two seats, but only one of us can sign in, and the login page keeps timing out.",
      "plan": "team"
    },
    "questions": {
      "queue": {
        "type": "choice",
        "instructions": "Which team should handle this message?",
        "criteria": {
          "billing": "Charges, invoices, refunds, seats on the bill",
          "technical": "Bugs, outages, login problems, integrations",
          "other": "Anything else"
        }
      },
      "angry": { "type": "noul", "instructions": "Is the customer angry?" }
    }
  }'

According to OpenRouter’s API reference, the response carries the same answers object you get from TypeSafe, plus a few OpenRouter fields:

Field	What it holds
answers	One typed answer per question: noul; choice with probabilities and confidence; score with legend, probabilities and confidence
id	A generation ID starting with gen-dec-
model	The dated version that answered, for example typesafe/jev-1.13-20260917
provider	TypeSafe
usage	input_tokens, output_tokens and cost in dollars

The usage.cost field is handy: in OpenRouter’s documented example, 476 input tokens cost $0.000019992, exactly 476 × $0.042 per million.

OpenRouter’s official TypeScript, Python and Go SDKs expose the same call under an Alpha.Decisions section.

Using TypeSafe’s official SDK with OpenRouter

OpenRouter also accepts TypeSafe’s own request and response format at POST https://openrouter.ai/api/v1/systemone. TypeSafe’s Python SDK documentation shows how to use it: set the base URL to https://openrouter.ai/api (or the environment variable TYPESAFE_BASE_URL), pass your OpenRouter key, and use the model ~typesafe/jev-latest.

typesafe_sdk_via_openrouter.py

import os

from typesafe_sdk import Noul, TypeSafeClient

with TypeSafeClient(
    api_key=os.environ["OPENROUTER_API_KEY"],
    base_url="https://openrouter.ai/api",
    model="~typesafe/jev-latest",
) as client:
    result = client.system_one(
        "I was charged twice.",
        {"billing": Noul(instructions="Is this about billing?")},
    )
    print(result.nouls["billing"].noul)

The rest of your code stays the same as on TypeSafe’s API, which makes this the easiest way to start on OpenRouter and move to TypeSafe later. The SDK’s model-listing call does not work against OpenRouter. We confirmed on September 25, 2026 that the route exists (it returns 401 without a key); we have not yet sent a paid request through it.

Differences from TypeSafe’s own API

• Context: 32,000 tokens on OpenRouter versus 64k per request on TypeSafe. In practice TypeSafe also caps the state plus the longest question at 32k, so large states hit a limit either way.
• Model naming: typesafe/jev-1.13 here; jev-1.13.0 or jev-latest on TypeSafe. The OpenRouter name will not work on TypeSafe, and TypeSafe rejects jev-1.13 without the patch number.
• Errors: OpenRouter wraps errors as {"error": {"code": ..., "message": ...}} instead of TypeSafe’s {"detail": ...}.
• SDKs: TypeSafe’s official SDKs work against OpenRouter’s /api/v1/systemone route with a changed base URL, as shown above. OpenRouter’s own SDKs use the Decisions endpoint.

Errors you are likely to see

HTTP	Message	What to do
401	“No cookie auth credentials found” (what we got with no header) or “Missing Authentication header” (documented)	Send Authorization: Bearer <key>
402	“Insufficient credits. Add more using https://openrouter.ai/credits”	Top up the account. See 402 Insufficient credits
400	“Invalid request parameters”	Check type, criteria and the model name
413	“Request payload too large”	Shrink the state
429	“Rate limit exceeded”	Back off and retry. See 429

We confirmed on September 19, 2026 that the Decisions endpoint exists (an unauthenticated POST returns 401 rather than 404). The response details above come from OpenRouter’s reference rather than our own logs, because we have not run paid calls through OpenRouter.

When OpenRouter is the right choice

OpenRouter is the quickest way to use Jev while TypeSafe’s own signups are paused: add credits and send the request above. If you already use OpenRouter for your LLM calls, it also keeps billing in one place. For long-term production use, compare it with the other channels, especially if you want the official SDKs or version pinning.