switchboardllm/switchboard-router-4b

TEXT GENERATIONPricing:Input $0.4 / Cached $0.08 / Output $0.8Concurrent Unit Cost:1Model Size:4BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Oct 13, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

The switchboardllm/switchboard-router-4b is a 4 billion parameter language model, fine-tuned from Qwen3-4B-Instruct-2507, designed for intelligent LLM routing. It analyzes user prompts and outputs a JSON object indicating the cheapest appropriate model tier (small, mid, or frontier) and domain (code, math, writing, general) to handle the request. This model's primary use case is cost optimization in LLM deployments by directing prompts to the most suitable and cost-effective backend model, improving upon its 0.5B predecessor by saving more and misrouting fewer hard prompts.

Loading preview...

switchboard-router-4b: LLM Routing for Cost Optimization

Switchboard-router-4b is a 4 billion parameter model, fine-tuned from Qwen3-4B-Instruct-2507, specifically engineered to act as an intelligent router for large language models. Its core function is to analyze incoming user prompts and determine the most cost-effective LLM tier capable of handling the request, outputting a single line of JSON. This allows developers to send simpler prompts to smaller, cheaper models while reserving expensive frontier models only for complex tasks that truly require them.

Key Capabilities

  • Intelligent Prompt Routing: Evaluates prompts and suggests the optimal model tier (small, mid, frontier) for execution.
  • Domain Classification: Provides a domain hint (code, math, writing, general) based on prompt keywords.
  • Cost Efficiency: Designed to significantly reduce operational costs by minimizing the use of high-cost frontier models.
  • Improved Performance: Outperforms its 0.5B predecessor, achieving higher cost savings (67.0% vs 64.5%) and reducing misrouted prompts (5.0% vs 5.5%) on a held-out dataset of 10,000 prompts.
  • Open Weights & Apache 2.0 License: Offers flexibility for local deployment and integration.

Good for

  • Optimizing LLM API Costs: Ideal for applications that interact with multiple LLM APIs of varying costs and capabilities.
  • Dynamic Model Selection: Automatically directs prompts to the most appropriate backend model based on complexity and domain.
  • Building Adaptive AI Systems: Enables the creation of more efficient and responsive AI applications by intelligently managing model resources.
  • Local Deployment: Available in GGUF format for running with Ollama or llama.cpp, and as a Transformers model for GPU inference.

Finetunes
2 models