---
title: "router-1: same answers, half the time, half the bill"
description: "router-1 picks the model and the effort for every call from the list you"
canonical: https://caveman.so/models/router-1
last-updated: 2026-10-18
---

# router-1: same answers, half the time, half the bill

router-1 picks the model and the effort for every call from the list you
allow: the frontier model for the hard work, something cheaper or a lower
effort for the rest. It is decision only: your call still goes to your
provider, on your keys.

- On 85 paired agentic coding tasks (SWE-Interact and SWE-Together), routing
  solved 43.7% against 44.3% for always running Opus 5.5, a
  difference inside the noise, at 0.46x the cost ($8.21 against $17.66 per
  solved task) and 2.2x faster (median wall time 15.9 against 35.2 minutes;
  1 agent timeout against 9). Interim run, list-price
  estimates, measured in Claude Code with an earlier router build. Most of
  the saving came from lower effort, not cheaper models.
- 4 ms p50 per decision.
- When it can't decide, you get your first model back with HTTP 200.

## Price

25,000 decisions free each month per organisation, then $0.001 a decision,
falling with volume to $0.0003. Published prices for the same decision,
October 2026: Morph Router $0.005, AWS Bedrock Intelligent Prompt Routing
$0.001 (inside Bedrock only), Not Diamond $0.05 per 1M tokens routed.

## API

```sh
curl https://gateway.caveman.so/v1/route \
  -H "Authorization: Bearer $CAVE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"models": ["claude-haiku-5-5", "claude-sonnet-5-5", "claude-opus-5-5"], "text": "Rename the helper and update its callers."}'
```

Body: `models` (1 to 64, yours, in your order) plus one of `text`,
`messages`, `signals` or `task`. Returns the model, the effort, a decision id
and a reason.

Get an API key: https://app.caveman.so/signup