clef-flash
by Cloudflare
The 9B Clef: typed decisions over text, images or video at about 39 ms median latency
Cloudflare/clef-flashDeploy clef-flash
Single-tenantMixpeek has no managed extractor for this model. On a single-tenant deployment you upload the weights and a custom plugin serves them next to the rest of your pipeline.
Overview
Clef-Flash is the 9B member of Cloudflare's Clef decision models. Like Clef, it takes a situation as text, JSON, images or video plus a set of typed questions, and returns a probability for every allowed answer in a single forward pass. Cloudflare released it on 30 September 2026 under Apache-2.0, post-trained from Qwen3.5-9B.
In Cloudflare's own Decision Index run it answers at a median of 38.8 ms against 209.3 ms for the 27B Clef, scores 90.9 macro-F1 on BANKING77 and 97.7 on a home appliance simulator. It is much weaker than Clef on CLINC150 with out-of-scope queries, 66.8 against 97.4.
The scores are self-reported.
Architecture
A Qwen3.5-9B backbone with its vision encoder reads the encoded record: the state, any images or video frames, and each question with its options. A small transformer head reads the backbone's final hidden states, routes evidence to each question and scores all options of all questions jointly, one logit per allowed option; a softmax per question gives the probabilities. It uses the same record format and API as the 27B Clef.
Mixpeek SDK Integration
# Ask typed questions about each item, then store the answers in metadata so
# Mixpeek search can filter or group by them.
import requests
for item in items: # [{"text": "...", "answers": {"department": "billing", "urgent": True}}]
requests.post(
"https://api.mixpeek.com/v1/buckets/bkt_your_bucket/objects",
headers={"Authorization": "Bearer API_KEY", "X-Namespace": "ns_your_namespace"},
json={
"blobs": [{"property": "body", "type": "text", "data": item["text"]}],
"metadata": item["answers"],
},
)Capabilities
- Typed questions: true/false, pick one of named options, or score on an ordered scale
- Reads the situation as text, JSON, images or video frames
- A probability for every allowed option of every question, in one forward pass
- Apache-2.0; same API as Clef, Jev and SystemOne
Use Cases on Mixpeek
Benchmarks
| Dataset | Metric | Score | Source |
|---|---|---|---|
| Decision Index | Median latency | 38.8 ms | Clef model card, Cloudflare's Decision Index 0.2.1 run (self-reported; Clef 209.3 ms) |
| BANKING77 | Macro-F1 | 90.9 | Clef model card (self-reported; Clef 94.2) |
| Home appliance simulator | Case exact accuracy | 97.7 | Clef model card (self-reported; Clef 83.0) |
| CLINC150 + out-of-scope | Macro-F1 | 66.8 | Clef model card (self-reported; Clef 97.4) |
Performance
Median latency 38.8 ms and p95 122.4 ms in Cloudflare's run. Weights are about 9.4B parameters with the vision encoder. We have not measured it.
Common Pipeline Companions
Frequently Asked Questions
What is the difference between Clef-Flash and Clef?
Clef-Flash is the 9B model, post-trained from Qwen3.5-9B; Clef is 27B. In Cloudflare's run Clef-Flash answers in 38.8 ms median against 209.3 ms and leads on tasks such as the home appliance simulator (97.7 against 83.0), while Clef is far stronger on CLINC150 with out-of-scope queries (97.4 against 66.8).
Can Clef-Flash read images and video?
Yes. It keeps the vision encoder of its Qwen3.5-9B backbone, so a record can include images or video frames alongside the text state.
Does Clef-Flash generate text?
No. It returns one probability per allowed answer for each question you ask, so there is no output to parse and no free text to check.
How do I use Clef-Flash answers in Mixpeek search?
Store each answer in the object's metadata and filter or group a retriever by it, as in the example on this page. Mixpeek taxonomies can also assign labels inside the pipeline.
Specification
Research Paper
Clef decision models (Cloudflare blog)
arxiv.orgBuild a pipeline with clef-flash
Add this model to a processing pipeline alongside other extractors. Combine with retrieval stages for end-to-end search.
Run it on your own data, free