Smart Route

A beta routing mode on LLM Gateway that classifies request difficulty and sends prompts to a cheaper or frontier model from a developer-selected pool.

LLM Gateway · Infrastructure and inference/serving · infrastructure

Published in Feed

Overview

LLM Gateway released Smart Route on September 25, 2026, a routing mode that lets developers name a pool of up to 30 eligible models and have a classifier rate each incoming request's difficulty, task type, and output type before picking a model from that pool -- unlike the platform's prior 'auto' selection, which had no visibility into task difficulty and no way to restrict which models it could pick. Requests can use no classifier (routes to the cheapest eligible model) or the 'Jev (TypeSafe)' classifier for difficulty-aware routing; session-based routing keeps multi-turn conversations on the same model, and the gateway returns an error rather than silently falling back when no approved model fits the request. It carries no platform fee during the beta and is available to all organizations, including pay-as-you-go, but is not yet available on DevPass.

This is a sourced directory entry. Features and capabilities are reported in the linked sources; this entry does not document independent Aymorphic testing.

Key features

  • Developer-selected pool of up to 30 eligible models
  • Optional Jev (TypeSafe) classifier rates request difficulty, task type, and output type
  • Session-based routing keeps multi-turn conversations on one model
  • Returns an error instead of a silent fallback when no approved model fits
  • Full logging of routing decisions: classifier used, difficulty rating, selected model

Use cases

  • Cutting inference cost by automatically routing simple prompts to cheaper models
  • Restricting which models a given application is allowed to resolve to
  • Auditing routing decisions via logged classifier and model-selection data

Pricing

no platform fee for Smart Route during beta; billed at standard per-model catalogue rates plus Jev classifier calls at jev-1.13.0 rates

Pricing reflects the source record and may change. Confirm current terms with the provider.

Dates and sources

Released
25 Sept 2026
Record updated
28 Sept 2026
Record last checked
28 Sept 2026

Explore the Feed

Other entries filed under infrastructure and inference/serving; inclusion does not imply equivalent features.

  • NVIDIA PAIR (Personal AI Router)

    NVIDIA's free, open-source personal AI router that spreads local model workloads across PCs on your network; out of beta Oct 13, 2026.

  • Agent37

    YC F26 startup selling low-cost agent sandboxes (gVisor or full VM) with prebuilt harnesses for Claude Code, Hermes and opencode.

  • Microsoft-Decision-1

    Microsoft's Qwen3.5-9B-based decision-scoring model returns calibrated option probabilities in one pass, at $0.042 per million input tokens.

Browse the Feed