---
title: "Clef Flash API hosted in the EU: typed decisions"
description: "Run Clef Flash on inference.eu: a 9B decision model that returns probabilities for typed questions instead of free text, processed in the EU."
canonical: "https://inference.eu/models/clef-flash/"
language: "en"
---

Chat · Cloudflare

# Run the Clef Flash decision model in the EU through our API

Clef Flash is a 9B multimodal decision model from Cloudflare. Instead of writing text, it reads a state, such as an email, a JSON record, an image or a video, and returns a probability for every allowed answer to a set of typed questions. On inference.eu you call it through our API, processed in EU data centres and never used for training.

[Contact sales](https://inference.eu/contact/)[Example request](#example)

1.  019B
2.  02EU data centres
3.  03Apache 2.0

## What Clef Flash is

Clef Flash was released by Cloudflare in October 2026 under the Apache 2.0 licence. It is post-trained from Qwen3.5-9B, keeps that model's vision encoder, and adds a small transformer head that scores answers rather than generating tokens. A larger variant, Clef, belongs to the same family.

You give the model a state and a schema of questions. Each question is one of three types: `noul` for true or false, `choice` for named options, or `score` for ordered options. In a single forward pass, the model returns one probability per allowed option for every question. There is no free-form generation, so there is nothing to parse and no risk of invalid output.

## What it does well, and where it falls short

Clef Flash suits the many decisions inside business software that are really classification: is this email a complaint, which team should handle the ticket, how urgent is it, does this image show damage. Because it answers every question at once and returns a probability for each option, you can set thresholds, route uncertain cases to people and log confidence for audits.

It is not a chat model. It cannot write replies, summarise, explain its reasoning or call tools, and it only answers the questions you define. Its probabilities are model estimates, so measure them against labelled examples before automating decisions. For open-ended tasks, use a chat model such as Qwen3.6 35B A3B or gpt-oss-20b next to it.

## Calling Clef Flash through the API

Clef Flash does not use chat completions. It uses the Jev and SystemOne decision format: send a `POST` request to `/v1/systemone` at the inference.eu base URL, with your inference.eu key and `model` set to `clef-flash`. The body holds the state, optional `instructions`, optional `images` or `videos`, and your questions with their `type` and `criteria`.

The response contains `answers` keyed by question ID. A `choice` answer gives the chosen option, a confidence and all probabilities; a `score` answer gives the expected score, a legend and probabilities; a `noul` answer gives the probability of true. Text-only and multimodal requests can share the same integration.

## Data residency for Clef Flash

Decision models usually read the raw material of daily operations: customer emails, claims, account records and photos. On inference.eu, Clef Flash runs on infrastructure in the European Union, so the state you send and the answers you receive are processed in EU data centres and are never used to train any model.

Every customer receives a data processing agreement under Article 28 of the GDPR and our subprocessor list. If an automated decision has legal or similarly significant effects on people, Article 22 GDPR still applies, so plan human review where needed. Billing is in euros. The Apache 2.0 licence allows commercial use without user limits.

## Clef Flash

Maker

Cloudflare

Made in

United States

Type

Chat

Parameters

9B

Licence

[Apache 2.0](https://huggingface.co/Cloudflare/clef-flash/blob/main/LICENSE)

Released

October 2026

Model ID

clef-flash

Processing

EU data centres

Training on your data

Never

Source

[huggingface.co](https://huggingface.co/Cloudflare/clef-flash)

clef-flash · Python

```
import os
from openai import OpenAI

client = OpenAI(base_url="https://api.inference.eu/v1", api_key=os.environ["INFERENCE_EU_API_KEY"])

response = client.chat.completions.create(
    model="clef-flash",
    messages=[{"role": "user", "content": "Summarise this contract in five bullet points."}],
)
print(response.choices[0].message.content)
```

## What teams use it for

-   1

    ### Ticket and email routing

    Decide the team, category and urgency of each incoming message in one call, with a probability for each option.

-   2

    ### Claims and document triage

    Check whether a submitted photo or document meets set criteria and send uncertain cases to a person.

-   3

    ### Content moderation

    Answer yes or no policy questions about text and images with confidence values you can threshold and log.

## Clef Flash: common questions

The questions buyers, developers and legal teams ask us most. Something missing? Ask us directly.

Compliance depends on how you process data. On inference.eu, Clef Flash requests are processed in EU data centres, not used for training and covered by a data processing agreement under Article 28 GDPR with a published subprocessor list. For automated decisions about people, also consider Article 22 GDPR.

A decision model answers predefined questions instead of writing text. Clef Flash reads a state and a schema of typed questions and returns a probability for every allowed option. It works like a flexible classifier that you configure with plain-language criteria rather than training data.

Three types: \`noul\` for true or false questions, \`choice\` for a set of named options, and \`score\` for ordered options such as a 0 to 4 severity scale. You can ask many questions of different types about the same state in one request.

No. Clef Flash has no free-form text generation. It only returns probabilities for the options you define. To write replies or summaries, pair it with a chat model: let Clef Flash decide, then let the chat model write.

Yes. It keeps the vision encoder of its Qwen3.5-9B base and accepts text, JSON, images and video as the state. You can ask, for example, whether a photo shows visible damage and how severe it is.

Yes. Clef Flash is released under the Apache 2.0 licence, following its Qwen3.5-9B base model. The licence allows commercial use, modification and redistribution without fees or user limits, including in paid products and internal tools.

A chat model can drift from the format, invent options and gives no reliable probability per option. Clef Flash scores every allowed option directly, so outputs are always valid and come with probabilities you can threshold, compare and monitor over time.

A reranker scores how relevant passages are to a query. Clef Flash answers your own typed questions about a state, with any number of options. Use a reranker such as bge-reranker-v2-m3 for search, and Clef Flash for classification, routing and checks.

Our API serves most models in the OpenAI format, but Clef Flash uses the Jev and SystemOne decision format at \`/v1/systemone\`, because its input and output are not chat messages. You use the same base URL and API key as for every other model.

Cloudflare developed Clef Flash by post-training Qwen3.5-9B from Alibaba's Qwen team. The weights are open. On inference.eu the model runs in EU data centres, so the data you send is not passed to Cloudflare or to Alibaba.

[Ask a question](https://inference.eu/contact/)

## Related

[All models](https://inference.eu/models/#directory)

-   [Qwen3.6 35B A3BRun Qwen3.6 35B A3B on inference.eu: a fast mixture-of-experts model with 3B active parameters, image input and 256K context, processed in the EU.Code](https://inference.eu/models/qwen3-6-35b-a3b/)
-   [gpt-oss-20bRun OpenAI's compact open-weight gpt-oss-20b reasoning model on inference.eu: the OpenAI API format, processed in EU data centres, billed in euros.Reasoning](https://inference.eu/models/gpt-oss-20b/)
-   [BGE Reranker v2 M3Run bge-reranker-v2-m3 on inference.eu: a light multilingual reranker that sorts search and RAG results by relevance, processed in the EU.Reranking](https://inference.eu/models/bge-reranker-v2-m3/)

## Move your AI workloads to EU infrastructure

Tell us which models you use and roughly how much traffic you have. We reply within one working day with pricing and next steps.

-   Reply within one working day
-   Pricing for your volume
-   DPA ready to sign