OpenRouter
A unified interface for LLMs — one OpenAI-compatible API in front of 400+ models across 70+ providers, with automatic fallback and price/performance routing.
Visit site →Microsoft Foundry Model Router
A trained routing model built into Azure AI Foundry that analyzes each prompt in real time and picks the most cost-effective model that meets a chosen quality bar.
Visit site →Cursor Router
Cursor's intelligent router for coding agents — a classifier trained on live requests that sends routine work to cheaper models and hard problems to frontier reasoning models.
Visit site →Ramp Router
One endpoint across leading closed and open models that automatically sends every request to the lowest-cost model clearing its quality bar, with built-in fallback and spend tracking.
Visit site →Sakana AI — Trinity
An ICLR 2026 paper introducing a lightweight, evolved coordinator that assigns Thinker, Worker, and Verifier roles across a pool of frontier LLMs — routing by role, not just by cost.
Visit site →Glean Auto Routing
Picks the optimal model from a pool of 40+ for every request based on evaluations across enterprise tasks, with cost-efficient, balanced, or frontier modes plus admin visibility into spend by user and department.
Visit site →Adaptive Router (ShipAI project)
A project write-up by Partha Sarkar on routing each step of a multi-agent pipeline to a different model: a low-cost Gemini Flash Lite classifier scores every step for complexity, reasoning and context, then picks a GPT-5 mini, GPT-5 or GPT-5 Pro tier, with reported savings of up to 40% versus all-GPT-5.
Visit site →Databricks Smart Routing (Unity Gateway)
A Beta feature of Databricks Unity Gateway that classifies each coding task up front with a small, fast model and routes it to a suitable model, or model plus harness via Omnigent, for the whole session to preserve cache hits. Works in Claude Code and Codex; Databricks reports 35% savings on its internal benchmark and 56% on public coding benchmarks (blog post, 13 August 2026).
Visit site →RouteLLM
Open-source framework for serving and evaluating LLM routers, which send easy queries to cheaper models to save cost. It accompanies the paper RouteLLM: Learning to Route LLMs with Preference Data. The repository's last commit was in August 2024, so check its activity before adopting it.
Visit site →vLLM Semantic Router
Open-source decision layer from the vLLM project for routing requests across models. GitHub lists Release v0.4.0 as the latest release, from September 2026.
Visit site →NVIDIA LLM Router blueprint
NVIDIA AI Blueprint for routing LLM requests to the best model for the task. GitHub lists v2.0.0-beta.1 (an experimental v2) as the latest release, from December 2025.
Visit site →LiteLLM routing
Router documentation for LiteLLM, the open-source gateway, covering load balancing, retries and fallbacks across deployments of the same or different models.
Visit site →Not Diamond
Model router for coding agents that, according to its site, predicts the best model for each input to cut cost without sacrificing accuracy.
Visit site →GitHub Copilot Fusion
GitHub Copilot's router and orchestrator, generally available from 28 October 2026: for each prompt it picks a single model, a cascade or a draft-and-critique pair, at a Standard, High or Max effort level. GitHub's own performance and cost claims are unverified here.
Visit site →