Breaking

Cloudflare and vLLM Semantic Router Release Open Decision Models to Challenge Jev

October 4, 2026 at 02:15 EDT

Foundation Models
AI Agents
Open Source

Two open-weight challengers to TypeSafe AI's Jev arrived within days, turning decision-only models from a single breakout idea into a fast-moving open category for AI agents. Cloudflare released Clef and Clef-flash under Apache 2.0, while the vLLM Semantic Router team published Decision 2.0, a six-model family built to answer structured questions about a request without generating prose.

Oct. 2026 · Open decision models

Decision models break open

Two open releases challenge Jev's early lead by turning agent control into typed probabilities instead of generated text.

Open rival releases

Decision 2.0 models

Questions per Clef request

13x

Clef-flash median speedup vs Jev

Latency shifts the hot path

Median request latency, shown as stacked vertical blocks scaled to about 40 ms each.

524.1 ms

Jev

209.3 ms

Clef

38.8 ms

Clef-flash

How the control layer works

State

→

Typed questions

→

Probabilities

→

Route or act

Why builders care

Bounded outputs and probabilities fit support triage, security scoring, tool-call checks, and model routing.

What remains unproven

Benchmarks need independent replication, and each deployment still needs thresholds, fallbacks, and domain-specific tests.

The likely split: decide fast, generate when needed

Clef and Decision 2.0 point toward agent stacks where a fast decision layer controls routing and guardrails while larger LLMs handle language, code, and deeper reasoning.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.