Breaking

AI/ML API Adds Jev Routing for Models and Tools in Its Agent MCP Connector

October 28, 2026 at 14:23 EDT

AI Agents
Foundation Models

AI/ML API has added Jev Router to its MCP offering for agents, using TypeSafe AI’s Jev to rate request difficulty and help select a suitable model and tool. The described routing pattern sends simpler tasks to lighter models such as GPT-6 Luna and harder reasoning work to Claude Opus 5.5.

AI/ML API · Agent routing

Jev routes each request to a fitting model and tool

A decision layer rates difficulty, directing routine work to lighter models and harder reasoning to stronger ones.

Same accuracy, lower test cost

One 205-task test reported 97.6% accuracy with Jev Router at 35% lower cost than Claude Opus 5.5.

100

Claude Opus 5.5 cost index

Jev Router cost index

Normalized test cost: Claude Opus 5.5 = 100; Jev Router = 65.

From request to response

Request

Task enters the agent flow →

Jev rates difficulty

A typed decision guides the route →

Choose model and tool

A fitting path is selected →

Response

The selected model generates

Potential benefit

Routine requests can use lighter models while harder work can reach stronger ones.

Trade-off

Decision overhead and model switching can reduce savings or cache reuse.

Continue reading

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