Switchboard Try the router

Open weights. Runs in your browser.

Pick the cheapest model
that can handle the prompt.

Most prompts don't need a frontier model. Switchboard is a 0.5B router that reads each prompt and sends it to a small, mid or frontier model, so you only pay frontier prices when the prompt needs it.

Route a prompt Get the weights Apache 2.0

0.5B parameters · one line of JSON out · about 15 tokens per decision

Real prompts from the held-out set and the tier Switchboard v0.1 actually picked. We show an equal number per tier, so the mix isn't to scale.

Three tiers, one decision per prompt

  • Small
  • Mid
  • Frontier

Try it on your own prompt

The real model runs in this tab. Nothing you type leaves your device.

The first route downloads the router once (660 MB). After that it runs offline in this tab.

How it works

Think of an old telephone switchboard. An operator listened to each caller and plugged the line into the right place. Switchboard does the same with your questions to AI: it reads each one and connects it to an AI that's good enough for it, without paying for the most expensive one every time.

1

You ask something

A question, a request to write or fix something, anything you'd type into a chatbot.

2

Switchboard reads it first

A small AI decides how hard the question is. "What's the capital of France?" is easy. A tricky math proof is hard. This takes less than half a second and costs almost nothing.

3

It goes to the right AI

Easy questions go to a small, cheap AI. Medium ones go to a mid-size AI. Only the hard ones go to the most powerful, most expensive AI.

4

You pay less for the same answers

Most everyday questions are easy, so most of them stop being billed at the top price. When Switchboard can't read its own answer, it plays it safe and picks the powerful AI.

Results

Loading the latest evaluation.

See the full evaluation

Up and running in a few minutes

Run it on your own computer. No GPU, account or API key needed.

  1. 1

    Get Ollama

    Download it free from ollama.com and open it once.

  2. 2

    Ask Switchboard

    Paste the line below into a terminal. The first run downloads the model once (994 MB).

  3. 3

    Read the answer

    The tier tells you which size of AI to use. Here, a small one is enough.

$ ollama run hf.co/switchboardllm/switchboard-router-0.5b-GGUF "Write a haiku about rain"
{"tier":"small","domain":"writing"}

From your own code

$ pip install switchboard-router

from switchboard_router import Router

router = Router()  # uses Ollama on this computer
router.pick_model("Write a haiku about rain")
'anthropic/claude-haiku-5.5'

No other dependencies. It works with Ollama, any OpenAI-compatible server, or Transformers, and falls back to the frontier model if routing fails. switchboard-router on PyPI

What's inside Switchboard

Size0.5 billion parameters. 994 MB to run on your computer, 660 MB in the browser.
Built onQwen2.5-0.5B-Instruct (Apache 2.0), fully fine-tuned
Trained on63,749 real prompts from routellm/gpt4_dataset (Apache 2.0)
Training time55 minutes on one RTX 5070 Ti
Answers withOne line of JSON: a tier (small, mid or frontier) and a topic (code, math, writing or general)
Runs onOllama, llama.cpp, Transformers, and your browser
LicenseApache 2.0, free for commercial use

Questions & answers

Switchboard

01What is Switchboard?

A small, open AI model that reads a prompt and decides how capable an AI needs to be to answer it: small, mid or frontier. You then send the prompt to an AI of that size.

02Why not always use the best AI?

Price. On Surplus, an answer from Claude Opus 5.5 costs 40 times more than the same answer from Claude Haiku 5.5. Most everyday prompts get an equally good answer from the smaller model.

03Which base model is it built on?

Qwen2.5-0.5B-Instruct from Alibaba, fully fine-tuned on routing examples. It's small on purpose: a router sits in front of every request, so it has to be fast and almost free.

04Who is behind it?

An independent project. It is not affiliated with Anthropic, OpenAI, Alibaba, Robinhood, Surplus or the Switchboard oracle network. Model names appear only to describe where prompts can be sent.

Model & use

05Does it run on my computer?

Yes. The 994 MB build runs with Ollama or llama.cpp on Windows, macOS and Linux, on a normal CPU. The demo on this page runs in your browser.

06Does my prompt leave my device?

Not in the demo on this page, and not when you run it with Ollama. If you call Switchboard through a hosted API, the prompt goes to that server like any other API call.

07How good is it?

On 10,000 test prompts it cut the bill by … compared with always using Claude Opus 5.5, and sent … of prompts to a cheaper tier than they needed. All the numbers, including the weak spots, are on the evaluation page.

08What if it picks the wrong tier?

Too high costs a bit more. Too low can mean a weaker answer, which is the mistake we work hardest to reduce. If its answer can't be read, it falls back to the frontier tier, so a broken routing call never sends you to a weaker AI.

09Is it free?

Yes. The model is Apache 2.0, and running it yourself costs nothing. After the Surplus launch you can also call it on the Surplus API at $0.03 per million tokens, about $0.000005 per routing decision.

10What data was it trained on?

Real prompts from routellm/gpt4_dataset (Apache 2.0, from LMSYS). Each prompt's label comes from how well a 2023 open model answered it, as scored by a GPT-4 judge. The model card has the details.

Coming to the Surplus API

After launch, call Switchboard on the same Surplus key as every other model, at $0.03 per million tokens.

Try the router Read the model card