Jev Router: OpenRouter Routing Model, Pricing & When to Use It

Jev Router vs Jev 1.13

The two names share a vendor and an OpenRouter organization page. They answer different questions. Router chooses which model speaks. Jev 1.13 scores the options you already defined.

How OpenRouter describes Jev Router

The model page says one endpoint gives you the model ecosystem, and that Jev reads the prompt to score difficulty and precision before the chat model runs. The launch thread adds four operational claims. Each one is OpenRouter's account of its own test or design, not a JevLab measurement:

• On four agent benchmarks, Jev Router solved 237 of 423 tasks, against 130 for OpenRouter's Auto Router.
• Median time to first token was faster than the other routers in that comparison.
• It keeps a working model for the rest of a session, and it changes effort without switching models when the gain is smaller than the cached chat it would lose.
• The Jev scoring call is described as zero data retention. Attachments are not sent to Jev. If the Jev call times out or returns invalid output, the request fails instead of falling back to another router.

TypeSafe's repost points people at the same listing. The organization page at openrouter.ai/typesafe shows three listings side by side: Jev Router, Jev Latest, and Jev 1.13.

jev-router.com is a different product. That site says it is independent of TypeSafe and does not provide TypeSafe's Jev. This page is about typesafe/jev-router on OpenRouter.

Is Jev Router free?

OpenRouter's Jev Router page FAQ says the listed price is zero, so that page states you are not charged for the router's prompt or completion tokens. The public models API records those prices as -1, the same variable-price marker OpenRouter uses for Auto Router, and Auto Router's FAQ uses the same zero-price sentence. The model the router selects can still be billed. Read the cost on the request.

Last verified: September 29, 2026, against the Jev Router page and the public models API. Auto Router uses the same -1 marker and the same FAQ sentence, and Auto Router still bills the model it selects. Treat "free" as the router page's reading of a non-fixed price.

A Dot post says Dot covers Jev Router's upstream cost on Dot's own platform. That offer is Dot's. It is not OpenRouter's list price.

Jev 1.13 has a published card: $0.042 per 1 million input tokens and $0 output, on both OpenRouter and TypeSafe's models page. Signup credits and which key you need are on Jev API access.

When to use Jev Router, and when to pin Jev 1.13

Use Jev Router when the product is a conversation, the acceptable models change by turn, and you want OpenRouter to choose the model and the effort. The 1,000,000-token figure on the listing refers to that router window. It does not raise Jev 1.13's context.

Pin jev-1.13.0 (OpenRouter id typesafe/jev-1.13) when the product is a label: routing, triage, an allow or deny gate, or any enum you need to retest at the same cutoff. Then set the cutoff yourself. The confidence threshold guide covers a three-band gate, and the threshold lab shows how auto-adopt and handoff move on one English five-label fixture. The slider starts at 0.80 as an interaction default, not a recommended production cutoff.

Theo Browne argued that Jev categorizes and does not reason, so it cannot tell how hard a coding task is from the prompt alone. He later published a DeepSWE run of Jev Router. Read those posts before you adopt OpenRouter's agent-benchmark claims, or before you treat the router as a stand-in for a pinned decision model. JevLab has not reproduced either test.

How to call Jev Router on OpenRouter

OpenRouter's page says the API is OpenAI-compatible and that the model slug is the part you change. Keep OPENROUTER_API_KEY on the server. This snippet is the documented shape, not a JevLab client:

curl https://openrouter.ai/api/v1/chat/completions \
  -H "Authorization: Bearer $OPENROUTER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "typesafe/jev-router",
    "messages": [
      { "role": "user", "content": "Summarize this diff in one paragraph." }
    ]
  }'

The full quickstart, including SDK base URLs, stays on the OpenRouter model page. For typed decisions, use the Decisions API on the Jev on OpenRouter page instead of this chat call. Key setup is on Jev API access. The lab's own TypeScript export is a five-label gate, documented at the docs, and it is not a client for typesafe/jev-router.

Sources checked on September 29, 2026

• OpenRouter: Jev Router
• OpenRouter: Jev 1.13
• TypeSafe models
• OpenRouter launch thread
• TypeSafe repost
• Theo Browne on System One and the DeepSWE note

Frequently Asked Questions