Cloudflare released Clef and Clef-flash on October 2, 2026, launching a set of fast decision models designed to route workflows and execute automated agentic decisions. Based on Qwen3.8-27B and Qwen3.5-9B, the models act as fast classification layers between heavy LLMs and simple code. Cloudflare reported median latencies of 39 milliseconds for Clef-flash and 209 milliseconds for Clef, directly undercutting competitor Jev’s 524-millisecond benchmark.

According to the company, these specialized models eliminate the need for human-in-the-loop verification for routine decisions by outputting precise structured probabilities. Clef supports a 64,000-token context window, vision processing capabilities, and complete API compatibility with TypeSafe AI’s Jev to simplify enterprise migration.

Alongside the model release, Cloudflare introduced a fine-tuning service using Reinforcement Learning for Calibrated Decisions (RLCD) to help customers adapt models to custom tasks. The launch signals a shift toward lightweight, low-latency decision layers that operate at edge data centers to trigger automated agent actions.

Why it matters

  • Replace slow, expensive LLM calls with 39ms decision models to lower latency and compute costs in agentic workflows.

  • Automate high-volume routing and security decisions using calibrated probability outputs rather than human verification.

  • Migrate seamlessly from Jev using Cloudflare’s drop-in compatible API and edge-hosted Qwen fine-tunes.

Source: the-decoder.com