Jev API Pricing: Free for a Limited Time

Jev is a System One decision model from TypeSafe. You send state and typed questions. It returns typed answers with probabilities. It does not write chat text. Founder Diogo Almeida's line, from the 16 September 2026 announcement: unstructured state in, typed probabilistic decisions out.

As of 5 October 2026, Jev is live on RouterPlex and free for a limited time. The model ID to call is jev-latest. The list price, which is what you pay when the promotion ends, is $0.042 per 1M input tokens. Output tokens are free either way. TypeSafe lists end-to-end latency at 70–500ms. The HTTP contract is POST https://api.routerplex.com/v1/systemone.

A call still needs a RouterPlex account with an active balance. A used-up balance returns 402, even while the promotion price is $0. These routes are not chat models, so do not paste jev-latest into Cursor or Claude Code and expect a coding agent.

Seven routes, one request shape

The three Clef IDs were added on 18 October 2026. They are Cloudflare's own decision models, open-weight under Apache 2.0 and served from Workers AI, and they are drop-in compatible with Jev: same state, same three question types, same answer fields. clef-flash is the one to reach for on a hot path; Cloudflare measured it 13x faster than Jev at the median, and its list price is below Jev's. clef is the 27B model for precision. The published prices are Cloudflare's Workers AI rates; RouterPlex adds nothing on top once the promotion ends.

jev-latest is the stable alias. jev-preview is the newest build. On 5 October 2026 both aliases resolve to jev-1.13.0. The response model field names the version that answered. Pin jev-1.13.0 once you tune confidence thresholds, so a later alias move does not shift the answers those thresholds were fit to. TypeSafe's /v1/models list omits jev-1.13.0. RouterPlex still accepts it.

decision-model-preview speaks the same request. No vendor publishes a list price for that preview, so $0.045 per 1M input is RouterPlex's cost, not a third-party sticker. It is free during the same promotion. Context was checked through the upstream on 5 October 2026: 64,014 tokens answered, and 70,000 tokens was rejected with a range of 1 to 65,536.

There is no end date on this page. The struck-through rates on the model index are the list prices after the promotion. Output stays free. Full request fields are in the decision model docs.

Not a smaller LLM

TypeSafe spent two years in stealth on a new architecture, a parallel sampler, and a training method they call Reinforcement Learning for Calibrated Decisions (RLCD). The AI primer puts that next to RLHF (chat) and RLVR (verifiable reasoning). Almeida helped build the instruction-following work behind ChatGPT; the docs call him a co-inventor of RLHF. The bet here is the opposite product: machine-to-machine decisions, not a better chatbot.

The names are on purpose. System One is Kahneman's fast System 1. Jev is Jevons: cheaper intelligence should increase demand, not just the margin.

It is also a narrower tool. Jev does not write replies, produce code, or explain its reasoning. If you needed a paragraph, you still need a chat model. If you needed a branch in code, this is the shape.

Three primitives

Every call is the same shape: a state (string, object, or array) plus a map of named questions. Questions are evaluated in parallel and in isolation against that state. Adding questions barely changes latency. TypeSafe's docs: keep each question to a snap judgment a knowledgeable person could make in a few seconds. Compose the rest in code.

A Noul near 0.5 is "I don't know," not "medium." Use a Score if you wanted a spectrum. Choice and Score confidence is a collapse of the probability distribution: peaked means sure, flat means escalate. TypeSafe's starting split is high / medium / low confidence → act / confirm / don't. The 0.5 floor is the "genuinely uncertain" cut; destructive actions sit higher.

On 16 September 2026, TypeSafe's materials put the request budget near 32,000 tokens for state plus questions together. The RouterPlex catalog on 5 October 2026 lists 64,000 for each Jev ID and 65,536 for decision-model-preview. That is still not a 1M chat window. Pack the state. Don't dump a repo.

JSON mode on GPT or Claude is not this. Those models still sample a string. You parse it, retry on schema errors, and hope the next turn doesn't invent a key. TypeSafe's system-one-adapter exists to force LLMs into this question/answer shape so you can compare cost and quality. The adapter is the LLM pretending to be TypeSafe, not Jev pretending to be ChatGPT.

Jev API pricing

TypeSafe's published list, 16 September 2026, and the RouterPlex bill on 5 October 2026:

The TypeSafe homepage also prints 238× lower input than Claude Fable 5.1. At $10 / 1M that arithmetic is exact ($10 / $0.042 ≈ 238). We list Claude Fable 5 at $10 / $50 (temporarily paused until further notice), not Fable 5.1. Don't treat 5 and 5.1 as the same row. The 238× figure is TypeSafe's comparison of list prices. It is not a RouterPlex benchmark, and it does not describe the promotion, which bills $0.

TypeSafe's own caveats, from the announcement: they cannot prove the price isn't subsidized; they expect it to go down, not up; published latency numbers were generally run from West Coast laptops, which is where the service currently sits.

A 10,000-token Jev state is $0.00042 of input at list price, and $0.00 during the promotion. Output does not add a line. That is why "ask extra speculative questions in the same call" is their default: the parallel-questions cookbook says batching 13 questions is about 12× cheaper and 10× faster than 13 separate calls, with the same answers.

When the promotion ends, the billed rate is the live catalog row, at list price, with no per-token markup. Prepaid balance. Hard per-key budget. The request stops at $0. Decision models are not on the chat rate card in the OpenRouter comparison. The same prepaid key still applies.

Workflow evals, not MMLU

TypeSafe did not lead with public chat benchmarks. They built workflow evals: assume a correct compute graph in code, ask the same narrow questions of every model, and score against a reference of GPT-6 Astra + Claude Fable 5.1 at high thinking. Four published workflows: security incidents, agent-trace observability, invoice processing, customer service.

Jev sits on the cheap side of that Pareto. The homepage's 193.6× faster / 444.6× cheaper figures come from this harness. TypeSafe says those numbers are on the higher end of real-world gains. The workflows were not in the training distribution, but they were written by TypeSafe's model-capabilities people, so some bias is possible. Using Astra and Fable 5.1 as the reference also pulls the target toward OpenAI and Anthropic.

Prompting the whole policy as one chain-of-thought, instead of the workflow, is worse for every model they plotted. Structure is the product.

That diagram is the point. Code owns close / queue / act. The model owns the readings. Grey-zone identity alerts notify a person. If you wanted Jev to "handle security," you would be using it wrong.

Type errors vs wrong answers

TypeSafe plots structured-output error rate and tool-call error rate at 0% for Jev. That 0% is not measured. Schema match is guaranteed, so they put a zero on the chart. LLM numbers in the same figure come from OpenRouter traffic, which they flag as biased toward harder queries on stronger models.

Read that chart as "it will not invent a key," not "it will not pick the wrong team." A Choice still returns one of your options. A Noul still returns a float. Calibration says that across many 0.8s, about 80% should be right. It does not certify the next ticket. Wire the confidence field (or the Noul itself) before you auto-close anything.

Jev specifications

Side demos TypeSafe published: a Doom bot on structured state (not pixels), about 10 queries/s, about $7/hour by their estimate; Wikiracing with high-cardinality link choice. Fun, not a product spec. Those dollar figures are TypeSafe's, at list price, not the promotion.

Call Jev on RouterPlex

This curl uses a RouterPlex key. Create one after you sign up and put a balance on the account. Swap jev-latest for jev-1.13.0, jev-preview, or decision-model-preview when you want that route. The Jev (latest) page shows the same call next to the live price.

curl https://api.routerplex.com/v1/systemone \
  -H "Authorization: Bearer $ROUTERPLEX_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "jev-latest",
    "state": "Help! My payouts have been failing for 3 days.",
    "questions": {
      "is_urgent": {"type": "noul", "instructions": "Does this convey urgency?"},
      "team": {
        "type": "choice",
        "instructions": "Which team should handle this?",
        "criteria": {"billing": "Payments, refunds", "technical": "Bugs, outages", "sales": "Pricing"}
      },
      "frustration": {
        "type": "score",
        "instructions": "How frustrated is the customer?",
        "criteria": ["Calm", "Frustrated", "Very angry"]
      }
    }
  }'

The TypeSafe Python SDK (tested with typesafe-sdk 0.7.2) works if the base URL is the host only. The SDK adds /v1. Pass the RouterPlex key. The JavaScript SDK takes the same root as baseURL.

import os
from typesafe_sdk import Choice, Noul, TypeSafeClient
 
client = TypeSafeClient(
    base_url="https://api.routerplex.com",
    api_key=os.environ["ROUTERPLEX_API_KEY"],
)
result = client.system_one(
    model="jev-latest",
    state="I was charged twice this month and nobody answers my emails.",
    questions={
        "billing": Noul(instructions="Is this a billing problem?"),
        "team": Choice(
            instructions="Which team?",
            criteria={"billing": "Payments", "technical": "Bugs"},
        ),
    },
)
print(result.answers["team"].choice, result.answers["team"].confidence)

Do not point an OpenAI chat client at this host and send messages. There is no chat adapter. Field-level notes are in the decision model docs.

When you still want a chat model

Jev will not draft the reply, write the patch, or explain the reasoning. Use it to branch. Use a chat model when the product needs sentences.

GLM-5.3 Flash is a cheap live chat route for classification you are willing to parse, and Kimi K3.1 Pro is live for harder judgment. Gemini 3.8 Flash, Claude Fable 5 at $10 / $50, and GPT-6 Astra, the flagship TypeSafe used as an eval reference, stay in the catalog but are temporarily paused until further notice. Volume chat routes, ranked by agent-turn cost, live on the cheapest models ledger. The OpenAI-compatible chat contract is a different page: one key, chat completions.

Create a RouterPlex account and call jev-latest on its model page. The promotion is free for a limited time. The list price is on that page for the day it ends.