Cloudflare Clef · Jev-compatible APIs
Clef vs Jev-compatible APIs
Clef and Clef-Flash are Cloudflare’s open multimodal decision models: 27B and 9B parameters, released on 1 October 2026. Clef-Omni followed on 9 October and also accepts audio. All three have Apache-2.0 open weights. Facts checked 27 October 2026.
Primary sources
Models and releases
Clef / Clef-Flash · 1 October 2026 · Clef-Omni · 9 October 2026. clef model card · clef-flash model card · clef-omni model card.
Prices and access
OpenRouter prices differ from Workers AI
USD per 1M uncached input tokens; output free. Checked 27 October 2026. The model pages do not establish an EU-only guarantee for these routes.
| Model / route | Input | Source |
|---|---|---|
| Clef · OpenRouter | $0.042 Prime Intellect; $0.240 Cloudflare; DeepInfra $0.192 (20% promotion, base $0.240) | OpenRouter |
| Clef-Flash · OpenRouter | $0.019 DeepInfra (50% promotion, base $0.038); $0.021 Prime Intellect; $0.038 Cloudflare | OpenRouter |
| Clef · Workers AI | $0.240 | Workers AI |
| Clef-Flash · Workers AI | $0.038 | Workers AI |
| Clef-Omni · Workers AI | $0.150 | Workers AI |
JevBench
Capability scores, method and run dates
JevBench and Benchmark Heaven are run by the company behind Decision Models.
| Model | Capability score | Run completed (UTC) | Measured in |
|---|---|---|---|
| Clef | 75.3 | 2026-10-04 | v1.6.0 |
| Clef-Flash | 63.8 | 2026-10-04 | v1.6.0 |
| TypeSafe Jev 1.13.0 | 77.1 | 2026-10-05 | v1.6.1 |
| TypeSafe Jev 1.14.0 | 75.0 | 2026-10-16 | v1.6.27 |
Source: Benchmark Heaven · JevBench · v1.7.11 aggregate JSON. Displayed method: v1.7 (10-second rule), release v1.7.11. Clef rows are self-hosted, measured on text; Jev uses its hosted API. The table names original measurement versions and completion dates; Clef was rescored in v1.7 from recorded timings. Capability averages Intelligence and Calibration; it is not an accuracy percentage. Different runs and access routes: these are not same-workload production measurements or speed claims. Our house models are not in Benchmark Heaven rankings. Facts checked 27 October 2026.
API calls
Similar JSON fields, different access
The Workers AI documentation shows state and a questions object with noul, choice and score. Workers AI Clef accepts 1–64 questions. Its account ID, Cloudflare token and model path replace a separate /v1/systemone endpoint. Check each provider’s media fields and limits rather than assuming an unchanged migration.
curl "https://api.cloudflare.com/client/v4/accounts/$CLOUDFLARE_ACCOUNT_ID/ai/run/@cf/cloudflare/clef" \
-H "Authorization: Bearer $CLOUDFLARE_AUTH_TOKEN" \
-H "Content-Type: application/json" \
-d '{"model":"clef","state":"Checkout is failing.",
"questions":{"urgent":{"type":"noul",
"instructions":"Is this request urgent?"}}}'curl https://api.decisionmodels.io/v1/systemone \
-H "Authorization: Bearer $DECISIONMODELS_API_KEY" \
-H "Content-Type: application/json" \
-H "S1-Region: eu" \
-d '{"model":"s1-fast","state":"Checkout is failing.",
"questions":{"urgent":{"type":"noul",
"instructions":"Is this request urgent?"}}}'Decision Models
Different models, EU tier
Decision Models does not serve Clef today. Our models are different models with a Jev-compatible request shape. If you need an EU-only tier with a public DPA for individual text decisions, you can evaluate s1-fast EU at €0.018 ($0.021) per 1M input tokens. If you need Clef itself or more questions per request, consider the Cloudflare routes above. A low input rate establishes neither better answers nor lower cost per completed task.
Jev is a product of TypeSafe AI; Clef is a Cloudflare model. OpenAI, Perplexity, Liquid AI, Microsoft and Bevel belong to their respective providers. Decision Models is not affiliated with them and uses the names only for comparison.