[ Guide ]
Jev on OpenRouter: model IDs and the Decisions API
Published By Jev Hub
[ Sponsors ]Launch price
These products help keep Jev Hub updated. Want yours up here?
- OpenRouter lists three Jev model IDs, typesafe/jev-1.14, typesafe/jev-1.13 and the alias ~typesafe/jev-latest, and you pay with an OpenRouter key and no separate TypeSafe account.
- It offers two ways in: its own Decisions API and a System One endpoint for TypeSafe's SDKs.
- OpenRouter also runs the Jev Router, which picks a model and reasoning effort per request, and publishes cookbooks for approving agent commands and checking cheap answers.
- Every figure from OpenRouter's cookbooks is its own.
Jev on OpenRouter, at a glance
| Feature | What to know | Where it is documented |
|---|---|---|
| Model IDs | typesafe/jev-1.14, typesafe/jev-1.13 and the alias ~typesafe/jev-latest | OpenRouter's Jev guide |
| Decisions API | POST https://openrouter.ai/api/alpha/decisions | OpenRouter's Jev guide |
| System One endpoint | POST https://openrouter.ai/api/v1/systemone, for TypeSafe's JavaScript or Python SDK with a changed base URL | OpenRouter's Jev guide |
| Key and billing | An OpenRouter key, billed to your OpenRouter account. No separate TypeSafe account or waitlist. Output tokens are free | OpenRouter's Jev guide |
| Price | No dollar figure on the guide: you pay per input token at the price on the Jev model page. Each response has a usage.cost field in US dollars | OpenRouter's Jev guide |
| Context | 262,144 tokens for state plus questions. More than about 32,000 bytes of text, and any image, runs in a slower, less accurate long-context mode | OpenRouter's Jev guide |
| Box questions | Need typesafe/jev-1.14 or the latest alias; Jev 1.13 answers them with a 400 error | OpenRouter's Jev guide |
Read on 25 October 2026 from OpenRouter's Jev guide, Jev Router page and cookbooks. The Decisions API path contains the word alpha, so check OpenRouter's current reference before you build on it.
How do I use Jev on OpenRouter?
Create an OpenRouter key and call one of the Jev model IDs. OpenRouter's guide says there is no separate TypeSafe account or waitlist, and usage is billed to your OpenRouter account. You pay per input token at the price on the Jev model page, and output tokens are free. The guide gives no dollar figure of its own, so read the model page for today's price.
A request has a model, a state and a questions object. Each question is a choice (pick one option), a score (a position on an ordered scale), a noul (a yes or no probability) or a box (a rectangle around something in an image). Answers come back with probabilities and a confidence value.
Which model ID should I use on OpenRouter?
Use typesafe/jev-1.14 for the newest version, typesafe/jev-1.13 if you need the older one, and the alias ~typesafe/jev-latest if you want to follow TypeSafe's newest release. The alias moves when a new version ships, so pin a version ID if you tune thresholds against it. Box questions work only on typesafe/jev-1.14 and the latest alias.
What is the difference between the two endpoints?
The Decisions API, POST https://openrouter.ai/api/alpha/decisions, is OpenRouter's own interface for decision models, with its own TypeScript, Python and Go SDKs. The System One endpoint, POST https://openrouter.ai/api/v1/systemone, speaks TypeSafe's format, so TypeSafe's own SDKs work if you change the base URL and use your OpenRouter key.
The guide never labels the Decisions API alpha or beta, but its path contains "alpha", and it calls image input and Box early features whose format, answers and limits can change between releases. Treat both as moving parts.
What is the Jev Router on OpenRouter?
The Jev Router, typesafe/jev-router, is a model ID that picks a model and a reasoning effort for each request. Jev judges the task and its difficulty, weighs the benefit of a stronger model against the estimated cost and your preferences, and the chosen model writes the reply. You can pick a cost tier, low, medium or high, and list models to prefer or exclude. It works with Chat Completions, Responses and Messages, and the response reports which model served it.
OpenRouter gives no price for the router itself, and says cost tiers are routing preferences, not a price cap. The estimated costs in its routing details are estimates, not final bills. It runs only in the global data region and cannot be combined with another router in one request.
What cookbooks does OpenRouter publish for Jev?
OpenRouter's Jev guide links five. Two have figures worth knowing, both OpenRouter's own and not independently checked:
- Auto-approve coding-agent permission prompts. A hook for Claude Code, Codex, Cursor and OpenCode sends safe-looking shell commands to Jev and approves them at a score of 0.9 or more, with a fixed risk list checked first. It measured about $0.0000168 per request.
- Cut LLM cost with a Jev-verified cascade. A cheap model drafts an answer from retrieved excerpts, Jev checks that the excerpts support it, and a stronger model runs only if the check fails, with an acceptance confidence of 0.8. On 50 questions it reports zero wrong answers at $0.012, against two wrong answers at $0.175 when the stronger model answered everything. The page says the cost is about 7% of the stronger model's, and warns that results varied by two answers between runs and that thresholds should be re-checked on your own data.
- Also listed: gating agent tool calls, classifying and tagging text at scale, and classifying Reddit and YouTube comments.
What goes wrong on OpenRouter?
The limits OpenRouter lists:
- Box on Jev 1.13. It returns a 400 error. Use the 1.14 ID or the latest alias.
- Images and long inputs run in long-context mode. OpenRouter says it is slower and less accurate than the normal mode. TypeSafe labels image and Box features as not ready for production.
- Box has no "not found" answer. Ask a
noulquestion in the same request to check whether the target exists. - The
stateformat is not settled on the page. Its Box example sends a list with a text part and an image part, while the text describes an object. Check the Decisions API reference before relying on a plain string.
Common questions
Yes. OpenRouter lists typesafe/jev-1.14, typesafe/jev-1.13 and the alias ~typesafe/jev-latest. You need only an OpenRouter key, and usage is billed to your OpenRouter account.
typesafe/jev-1.14 for the newest version, typesafe/jev-1.13 for the older one, and ~typesafe/jev-latest for an alias that follows the newest release.
POST https://openrouter.ai/api/alpha/decisions. For TypeSafe's own SDKs, OpenRouter also offers POST https://openrouter.ai/api/v1/systemone.
No. OpenRouter's guide says there is no separate TypeSafe account or waitlist. Usage is billed to your OpenRouter account.
A model ID, typesafe/jev-router, in which Jev picks the model and reasoning effort for each request, with low, medium and high cost tiers. OpenRouter gives no price for the router itself.
The context length is 262,144 tokens. OpenRouter says requests with more than about 32,000 bytes of text, and every request with an image, run in a slower, less accurate long-context mode.
Yes, as an early feature on Jev 1.14 and 1.13, in long-context mode. Box questions, which return a rectangle around an object, need Jev 1.14 or the latest alias.
Sources
Written by Jev Hub from the sources below and checked against them on . Where a figure is a company's own claim, the article says so.
- OpenRouter: using Jev through OpenRouter (read 25 October 2026) (opens in a new tab)
- OpenRouter: Jev Router (read 25 October 2026) (opens in a new tab)
- OpenRouter cookbook: auto-approve coding-agent permission prompts with Jev (read 25 October 2026) (opens in a new tab)
- OpenRouter cookbook: cut LLM cost with a Jev-verified cascade (read 25 October 2026) (opens in a new tab)
- TypeSafe docs: models and limits (read 25 October 2026) (opens in a new tab)
Related content
- Read the article
Guide
Published
Confidence routing
Use Jev's confidence number to decide when to act, ask or hand off.
- Read the article
Guide
Published
Intent routing
Let Jev read each request and pick who handles it: code, a model or a person.
- Read the article
Guide
Published
Fan-out
Ask every question in one request and use only the answers that apply.