Clef Models by Cloudflare

CloudflareApache 2.0Open sourceAgents
Cloudflare releases · 5 in the last 12 months · this family litChangelog →
3 models2026Up to 256K ctxFrom $0.038/1M input

Last refreshed 2026-10-15. Next refresh: weekly.

Details

ResearcherCloudflare
LicenseApache 2.0OSI-approved
Commercial usePermitted
Models3
Released2026
Max context256K

Capabilities

VisionAll models
MultimodalAll models
Structured OutputsAll models

About

Clef is Cloudflare's family of open-weight multimodal decision models, trained by the Workers AI team. A Clef model reads a state plus a schema of typed questions (choice, yes/no (called "noul" in the API) or score) and returns a probability for every allowed option in one pass, with no free-form text generation. Three models ship on Workers AI: Clef (27B, post-trained from Qwen3.8-27B, 64K context) for highest-precision decisions; Clef-flash (9B, post-trained from Qwen3.5-9B) for latency-critical hot-path decisions, with a 24K hosted context since its 2026-10-09 price cut; and Clef-omni (30B mixture-of-experts with 3B active, post-trained from Qwen3-Omni-30B-A3B-Instruct, 64K context), added 2026-10-09, which also reads audio and video. Clef and Clef-flash launched 2026-10-01 and accept text, JSON, and up to four images per request. All three bill media as input tokens, do not charge for output tokens, follow the System One API so Jev integrations can switch by changing endpoint and model, and are published on Hugging Face.

Decision facts

Best fit
agentsvision and multimodal workstructured outputs
Capability starting point
Clef-omni with 64K context and structured outputs and multimodal inputs
Lowest tracked input
Clef-flash · $0.038/1M · Cloudflare Workers AI
Closest related family
Agents-A1

Current Variants

Use-when guidance is based on each model's tracked capabilities, context window, release date, and replacement status.

3 in view
Clef-omniCurrent

Use when the workload needs 64K context, structured outputs, and multimodal inputs.

2026-1064K contextstructured outputsmultimodal inputs
ClefCurrent

Use when the workload needs 64K context, 27B parameters, and structured outputs.

2026-1064K context27B parametersstructured outputs
Clef-flashCurrent

Use when the workload needs 256K context, 9B parameters, and structured outputs.

2026-10256K context9B parametersstructured outputs

Release Timeline

1 release group
2026-10
3 current
Clef
64K context27B parametersstructured outputs
Current
Clef-flash
256K context9B parametersstructured outputs
Current
Clef-omni
64K contextstructured outputsmultimodal inputs
Current

Specifications(3 models)

Clef model specifications comparison
ModelReleasedContextParametersVisionMultimodalStructured Outputs
Clef-omni2026-1064K30B total / 3B activeYesYesYes
Clef2026-1064K27BYesYesYes
Clef-flash2026-10256K9BYesYesYes

Available From(1 provider)

Pricing

Clef model pricing by provider
ModelProviderInput / 1MOutput / 1MType
Clef-flashCloudflare Workers AI$0.038—Serverless
Clef-omniCloudflare Workers AI$0.15—Serverless
ClefCloudflare Workers AI$0.24—Serverless