Zuse, the European Jev alternative.Same requests. Open today.

Zuse is a decision model like TypeSafe's Jev: it reads a document, picks from the options you set and tells you how sure it is. Point your Jev code at Zuse and it runs, hosted in Sweden.

Get an API keyCompare on your data

Or try Zuse in the playground, no account needed.

.env
TYPESAFE_BASE_URL=https://api.typesafe.ai
TYPESAFE_BASE_URL=https://api.bineric.com
TYPESAFE_API_KEY=your-typesafe-key
TYPESAFE_API_KEY=your-bineric-key
TYPESAFE_DEFAULT_MODEL=jev-latest
Your code: unchanged. jev-latest is answered by Zuse 27B.

What stays the same, and what changes

Stays the same

  • The request and the response
  • TypeSafe's Python and JavaScript SDKs
  • Choice, yes/no and score questions
  • A confidence with every answer
  • Jev model names, such as jev-latest

Changes

  • The base URL and the API key
  • Hosted in Sweden, with a data processing agreement
  • 115 languages, 27 European trained
  • Image (box) questions are not supported

Same request, same answer

Send the request you send Jev today to /v1/systemone. The answer comes back in Jev's format: a choice, a yes/no probability or a score, each with how sure it is.

Your Jev request
POST https://api.bineric.com/v1/systemone


{
  "model": "jev-latest",
  "state": "Hi, I was charged twice for my monthly pass last week. I want my money back as soon as possible, this is the second time it has happened!",
  "questions": {
    "team": {
      "type": "choice",
      "instructions": "Which team should handle it?",
      "criteria": {
        "billing": "Payments and refunds",
        "technical": "The app or website is not working",
        "travel": "Questions about routes and departures",
        "other": "Other"
      }
    },
    "refund": {
      "type": "noul",
      "instructions": "Is the customer asking for a refund?"
    },
    "urgency": {
      "type": "score",
      "instructions": "How urgent is it?",
      "criteria": [
        "Not urgent",
        "Normal",
        "Urgent",
        "Critical"
      ]
    }
  }
}
Zuse's answer
{
  "model": "zuse-27b",
  "answers": {
    "team": {
      "type": "choice",
      "choice": "billing",
      "confidence": 0.9912,
      "probabilities": {
        "billing": 0.9912,
        "technical": 0.0031,
        "travel": 0.0009,
        "other": 0.0048
      }
    },
    "refund": {
      "type": "noul",
      "noul": 0.9874
    },
    "urgency": {
      "type": "score",
      "score": 2.3001,
      "legend": {
        "0": "Not urgent",
        "1": "Normal",
        "2": "Urgent",
        "3": "Critical"
      },
      "probabilities": {
        "0": 0.0102,
        "1": 0.0915,
        "2": 0.4863,
        "3": 0.412
      },
      "confidence": 0.4863
    }
  },
  "usage": {
    "input_tokens": 214,
    "output_tokens": 0
  },
  "latency_ms": 18.6
}

Zuse and Jev side by side

On Jev's own public English benchmark, Zuse matches its accuracy and can safely automate more decisions. Jev is ahead on general knowledge and reads images.

MeasureZuse 27BJev (published)
Decisions safe to automate (errors under 5%)77.6%70.0%
Wrong answers given with high confidence1.7%3.7%
Accuracy86.0%85.7%
Rule-and-evidence reading (SemIf)95.6%96.5%
General-knowledge questions (MMLU-Pro)about 60%84%
Image questionsNoYes
Languages115 languages, 27 European trainedNo list published

Jev's figures are the ones TypeSafe published. Zuse was scored once on the frozen test split of Jev's public English suites and never tuned on it (21 October 2026). Bold marks the one ahead.

Run both on your own data

Benchmarks are a start. Your documents are what count. Send the same ones to Jev and Zuse and see where they agree.

  1. Put your documents in documents.txt, one per paragraph.
  2. Set TYPESAFE_API_KEY and BINERIC_API_KEY.
  3. Run it, and change the question to one of yours.
compare.py
import os, requests


APIS = {
    "Jev":  ("https://api.typesafe.ai", os.environ["TYPESAFE_API_KEY"]),
    "Zuse": ("https://api.bineric.com", os.environ["BINERIC_API_KEY"]),
}
QUESTIONS = {"team": {
    "type": "choice",
    "instructions": "Which team should handle it?",
    "criteria": {"billing": "Payments and refunds",
                 "technical": "The app is not working",
                 "other": "Other"},
}}


# One document per paragraph
for doc in open("documents.txt").read().split("\n\n"):
    got = {}
    for name, (base, key) in APIS.items():
        r = requests.post(f"{base}/v1/systemone", timeout=30,
            headers={"Authorization": f"Bearer {key}"},
            json={"model": "jev-latest", "state": doc, "questions": QUESTIONS})
        got[name] = r.json()["answers"]["team"]
    jev, zuse = got["Jev"], got["Zuse"]
    same = "same" if jev["choice"] == zuse["choice"] else "DIFFERENT"
    print(f"{same:9}", f"Jev {jev['choice']} {jev['confidence']:.0%}",
          f"Zuse {zuse['choice']} {zuse['confidence']:.0%}", doc[:40])

Made for European companies

What you need before a decision model can touch customer data.

Hosted in Sweden

Requests are processed in memory on servers in Stockholm. Documents and questions are not stored, not logged and never used for training.

GDPR paperwork done

A data processing agreement applies automatically, and the list of subprocessors is public. Nothing to negotiate before you start.

Built for 115 languages

Trained on 27 European languages, with the score for every language published, from Norwegian to Maltese.

Your own instance

Reserve a GPU in an EU data centre for your traffic alone, run Zuse in your own AWS account from AWS Marketplace, or on your own servers.

Read the data processing agreement and the list of subprocessors.

Questions

Is Zuse made by TypeSafe, or built on Jev?

No. Zuse is built and trained by Bineric in Norway. No output from Jev was used to train it. It accepts the same request format as Jev so that code written for Jev runs on it.

Does our Jev code work unchanged?

For choice, yes/no and score questions, yes: set the base URL to https://api.bineric.com and use a Bineric API key. Requests to /v1/systemone that name a Jev model, such as jev-latest, are answered by Zuse 27B. TypeSafe's Python and JavaScript SDKs work as they are. Image (box) questions are not supported.

Can we try both and compare?

Yes. Because the requests are the same, you can send the same documents to both and compare the answers on your own data before you decide.

How do we get access?

Sign up, add credit and create an API key. There is no waitlist. You can also try Zuse without an account in the playground on the Zuse page.

Start with Zuse today

Get a key, point your code at Bineric, and the answers come straight back.

Get an API keyMore about Zuse

Jev and TypeSafe are trademarks of their owner. Bineric is not affiliated with TypeSafe.