Cloudflare Clef
by Cloudflare
Cloudflare Clef is a multimodal decision-model family launched October 1, 2026, designed for structured probabilistic choices rather than free-form text generation. Clef (27B) and Clef-flash (9B) return probability distributions over allowed choices, suited for routing, escalation, threat classification, and agent guardrail decisions. Weights are available under Apache 2.0; models are hosted on Cloudflare Workers AI.
Primary Use Cases
- Agent routing and escalation decisions
- Threat classification and security guardrails
- Agent 'should I act?' decision gates
- Structured classification tasks in production pipelines
✅ Strengths
- Purpose-built for structured decisions rather than general text generation
- Apache 2.0 open weights enable self-hosting
- Integrated with Cloudflare Workers AI for edge deployment
- Low latency Clef-flash variant for real-time use
⚠️ Limitations
- Benchmark results are vendor-reported; independent evaluation pending
- Latency and accuracy claims require independent verification
- Not a general-purpose language model — limited to structured decision tasks
🏆 Performance Benchmarks
Standardized benchmark scores for objective comparison
Vendor-reported; Cloudflare claims Clef leads 7 of 10 evaluated decision-model categories. Independent evaluation pending.
Vendor-reported median latency for Clef-flash 9B model. Independent verification pending.
Technical Specifications
Performance Benchmarks
Source: View benchmark details
Trust & Privacy
Ecosystem & Integrations
Pricing Breakdown
Hosted inference via Cloudflare Workers AI; free tier available under Workers AI free plan
Open weights available under Apache 2.0 license for self-hosting
0