Independent measurements · not TypeSafe AI · we do not sell keys
Jev API key and access while TypeSafe signup is paused
TypeSafe paused new direct signups on 22 September 2026. You can still use Jev today with an OpenRouter, Vercel, Netlify or Cloudflare key, no TypeSafe account required. OpenRouter is the route we tested end to end.
- Get a key now: create an OpenRouter account and add credit
- Endpoint:
POST openrouter.ai/api/v1/systemone - Price: $0.042 per 1M input tokens; output is free
- Context: about 32K on every gateway (64K direct); each call bills ~260 extra tokens
Which way in suits you?
The fastest route we have verified: an OpenRouter key with credit on it, then run the call below.
- Key from
- OpenRouter
- Model ID
- typesafe/jev-1.13
- Price
- $0.042 / 1M input, output free
- Context
- 32K
curl https://openrouter.ai/api/v1/systemone \
-H "Authorization: Bearer $OPENROUTER_API_KEY" \
-H "Content-Type: application/json" \
-d '{"model":"typesafe/jev-1.13","state":"I was charged twice for March. Please refund one.","questions":{"refund":{"type":"noul","instructions":"Does the customer want a refund?"}}}'Tested: 3,000 requests on 23 Sep 2026, billed at exactly tokens × $0.042/1M. Checked .
What comes back, what it costs, what can go wrong
Reading the request
state is the text you ask about. Each entry in questions is one decision, and noul asks a yes/no question. The answer comes back keyed by the same name, as a probability:
{
"answers": {
"refund": {
"type": "noul",
"noul": 0.97
}
},
"usage": {
"input_tokens": 284
}
}The shape of a response; your probability will differ.
What it costs
This exact call bills about 284 tokens, most of it the fixed per-call overhead. A thousand of them cost about $0.012; a million, about $11.93.
Estimate your own request →Monthly cost of a workload
If it fails
- 400
- The body is wrong, or the request is too big: we saw 400 for every request past the context limit.
- 401
- The key is missing or not valid for this route.
- 402
- No credit on the account. OpenRouter and Cloudflare both answered 402 until credit was added.
- 429
- Rate limit: 250,000 tokens/sec or 1,200 requests/min direct. Back off and honour retry-after.
Every route side by side
- OpenRouterNo TypeSafe account
32K context · $0.042 / 1M input, output free
typesafe/jev-1.13
Tested: 3,000 requests on 23 Sep 2026, billed at exactly tokens × $0.042/1M.
- TypeSafe directNeeds TypeSafe key
64K total, 32K state + longest question context · $0.042 / 1M input, output free
jev-latest (= jev-1.13.0)
Not tested here: signup was closed when we tried.
- Vercel AI GatewayNo TypeSafe account
32K context · $0.042 / 1M input, output free
typesafe-ai/jev
Not tested here.
- Netlify AI GatewayNo TypeSafe account
~32K context · Billed in Netlify credits (no Jev rate published)
jev-latest (= jev-1.13.0)
Not tested here.
- Cloudflare Workers AINo TypeSafe account
32K context · $0.042 / 1M input, output free
typesafe/jev
Tested: requests authenticate, then return 402 until AI Gateway credits are loaded.
- LiteLLMNeeds TypeSafe key
TypeSafe's (64K) context · TypeSafe's rate (it forwards)
jev-latest (passed through)
Not tested here: it forwards to TypeSafe, so it still needs a TypeSafe key.
Also listed, not included: Requesty shows Jev from $0.04 / 1M input plus a 5% pay-as-you-go fee, but its sample calls it through chat completions, which a decision model does not use. We have not verified it works.
The 32K gateways will refuse what 64K direct accepts
Direct, a request can hold up to 64K tokens in total; in our tests the cut-offs sat at 32,768 tokens for the state plus the longest question and 65,536 in total. Every gateway lists about 32K. If your state is long, or you batch many questions into one call, check the size first: paste your request into the token estimator.
What a call actually costs in time and tokens
| State size | Tokens billed | Latency (warm) |
|---|---|---|
| 1,000 tokens | 1,286 | 375 ms |
| 4,000 tokens | 4,317 | 382 ms |
| 8,000 tokens | 8,354 | 478 ms |
| 16,000 tokens | 16,429 | 571 ms |
| 24,000 tokens | 24,502 | 571 ms |
| 32,000 tokens | 32,579 | 734 ms |
- Reuse the connection. The same small request took 1,630 ms when it had to open a new connection first, against 375 ms on a reused one. Rate limits and latency.
- Ask several questions per call. Every call bills ~260 extra tokens on top of your text and questions, so one call with four questions costs far less than four separate calls. How much less.
- No hidden charges seen. In our OpenRouter test the bill matched tokens × $0.042 / 1M, with nothing extra.
Jev API questions
How do I get a Jev API key now that TypeSafe signups are paused?
Use a gateway: OpenRouter, Vercel AI Gateway, Netlify AI Gateway, Cloudflare Workers AI all serve Jev without a TypeSafe account, with a key from the gateway itself (Netlify injects it for you). TypeSafe paused new direct signups on 22 September 2026; existing TypeSafe accounts keep working.
Which endpoint and model ID do I use?
Direct: POST https://api.typesafe.ai/v1/systemone with model jev-latest. OpenRouter: POST https://openrouter.ai/api/v1/systemone with typesafe/jev-1.13, the same body. Vercel: typesafe-ai/jev through the AI SDK's evaluate API. Cloudflare: typesafe/jev through Workers AI.
Is the Jev API price the same on every route?
Yes where a rate is published: TypeSafe, OpenRouter, Vercel and Cloudflare all list $0.042 per 1M input tokens with output free. Netlify bills Jev in Netlify credits and publishes no separate rate. Each request also bills about 260 tokens of overhead on top of your state and questions.
Do the gateways give the same context window as TypeSafe?
No. TypeSafe documents 64K tokens per request (32K for the state plus the longest question). OpenRouter, Vercel, Cloudflare and Netlify list about 32K. A request that uses more than 32K in total works direct but not through those gateways.
What do Jev API errors 400, 401, 402 and 429 mean?
400: the body is malformed or the request is over the context limit (we saw 400 for every over-limit request). 401: the key is missing or invalid for that route. 402: the account has no credit; OpenRouter and Cloudflare both returned it until credit was added. 429: a rate limit; back off and honour retry-after.
How fast is the Jev API?
In our test through OpenRouter, a request with 1,000 tokens of state took 375 ms on a warm connection and 32,000 tokens took 734 ms; the same small request on a fresh connection took 1,630 ms, handshake included. Reuse connections.
Is there a free Jev API?
Not an ongoing one. Vercel's launch promotion made Jev free on AI Gateway until 25 September 2026; its model API now lists $0.042 per 1M input. Accounts opened during TypeSafe's open signup got $5 of credit, reported by VentureBeat.
Get an email when TypeSafe reopens direct signup
TypeSafe gave no reopening date. Leave an address and we will tell you when you can get a key direct again.
Routes and rates from each route’s own page, checked 2026-09-26; the signup pause is TypeSafe’s post. Measurements: jev-fanout-bench. 2 of 6 routes were called by us; the rest are marked. This site is independent of TypeSafe AI and does not sell API access.