Stay updated with the latest in technology, global innovations, and key economic trends. From AI breakthroughs to global energy market insights, we bring you the news that matters.
Cloudflare Unveils Clef Qwen-Powered AI Decision Engine Built for High-Speed Routing.
Get link
Facebook
X
Pinterest
Email
Other Apps
-
Cloudflare Launches Clef AI Decision Engine Powered by Qwen Architecture to Lower Enterprise Agentic Routing Costs
Cloudflare, Inc. has officially expanded its specialized artificial intelligence portfolio with the debut of Clef, a high-speed decision-making and routing AI model. Building upon the framework established by Jev Cloudflare’s previous decision engine launched in mid-September Clef replaces the original underlying base architecture with Qwen open-weight models, offering enterprise teams near-parity performance at significantly reduced operational inference costs.
Model Variants, Speculative Prefill Architecture, and Economic Specifications
Dual-Tier Architecture Built on Qwen Foundation:
Standard Clef Model: Built using Qwen 3.8-27B as its foundational core, engineered for complex decision logic, policy evaluation, and intricate workflow routing.
Clef-Flash Model: Powered by Qwen 3.5-9B, optimized for ultra-low latency decision tasks, high-throughput payload evaluation, and lightweight edge routing.
Novel Prefill-Only Inference Engine:
Decoding Elimination: Unlike standard large language models (LLMs) that execute slow token-by-token auto-regressive decoding, Clef-flash operates strictly during the prefill phase.
Schema-Constrained Outputs: The model completely bypasses standard text decoding. Instead, output responses are restricted to predefined structural choices and JSON schema options. This enables instantaneous, deterministic output selection without the computation overhead of text generation.
Benchmarking Performance & Cost Breakdown:
Competitive Decision Parity: Internal benchmarks demonstrate that Clef matches the decision-making precision of Jev across core evaluation suites, trailing slightly only in long-horizon agent trace reasoning benchmarks.
Highly Competitive Pricing Structure: Standard Clef is priced at $0.24 per million tokens, while the lightweight Clef-flash variant operates at $0.09 per million tokens, drastically undercutting general-purpose commercial reasoning APIs.
Ask me anything about this article. No data is stored for your question.
Cloudflare Launches Clef AI Decision Engine Powered by Qwen Architecture to Lower Enterprise Agentic Routing Costs
Cloudflare, Inc. has officially expanded its specialized artificial intelligence portfolio with the debut of Clef, a high-speed decision-making and routing AI model. Building upon the framework established by Jev Cloudflare’s previous decision engine launched in mid-September Clef replaces the original underlying base architecture with Qwen open-weight models, offering enterprise teams near-parity performance at significantly reduced operational inference costs.
Model Variants, Speculative Prefill Architecture, and Economic Specifications
Dual-Tier Architecture Built on Qwen Foundation:
Standard Clef Model: Built using Qwen 3.8-27B as its foundational core, engineered for complex decision logic, policy evaluation, and intricate workflow routing.
Clef-Flash Model: Powered by Qwen 3.5-9B, optimized for ultra-low latency decision tasks, high-throughput payload evaluation, and lightweight edge routing.
Novel Prefill-Only Inference Engine:
Decoding Elimination: Unlike standard large language models (LLMs) that execute slow token-by-token auto-regressive decoding, Clef-flash operates strictly during the prefill phase.
Schema-Constrained Outputs: The model completely bypasses standard text decoding. Instead, output responses are restricted to predefined structural choices and JSON schema options. This enables instantaneous, deterministic output selection without the computation overhead of text generation.
Benchmarking Performance & Cost Breakdown:
Competitive Decision Parity: Internal benchmarks demonstrate that Clef matches the decision-making precision of Jev across core evaluation suites, trailing slightly only in long-horizon agent trace reasoning benchmarks.
Highly Competitive Pricing Structure: Standard Clef is priced at $0.24 per million tokens, while the lightweight Clef-flash variant operates at $0.09 per million tokens, drastically undercutting general-purpose commercial reasoning APIs.
Comments
Post a Comment