Breaking

Developers Pair Claude Opus 5.5 With TypeSafe Jev for Adaptive Effort Routing in Claude Code

October 8, 2026 at 02:27 EDT

Foundation Models
AI Agents
Software Dev & Coding

Developers are experimenting with a new Claude Code setup that combines Anthropic's Claude Opus 5.5 with TypeSafe Jev, using Jev as a fast decision layer that re-selects reasoning effort before each model call. The approach is packaged in the MIT-licensed jev-opus project, which adds an "Opus 5.5 · Jev" model entry inside Claude Code and routes requests through a local gateway while preserving Anthropic prompt caching.

Oct. 8, 2026 · Claude Code

Opus 5.5 gets a Jev router

A local gateway lets TypeSafe Jev re-pick Claude Opus 5.5 effort before each call, aiming to keep hard work on Opus while preserving prompt-cache gains.

Opus 5.5 input per million tokens

$20

Opus 5.5 output per million tokens

0.26s

Jev p50 latency in one Banking77 run

Routing cost comparison

Cost per 1,000 decisions in one 154-message Banking77 benchmark.

$13.68Opus 5.5 only

$3.33Jev then Opus

The routed setup stayed within 0.7 percentage points of Opus-only accuracy in that run.

How the loop works

1. Request

Claude Code sends the next turn.

→

2. Jev decides

Effort is selected for the step.

→

3. Opus acts

Claude Opus 5.5 handles the work.

Why developers like it

It targets lower cost, keeps the Claude Code interface, and avoids losing cache when effort changes.

What remains uncertain

The public benchmark is a single small run, and some Claude surfaces are unsupported or untested.

Continue reading

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