How to Use Claude Code with OpenRouter

TL;DR Learn how to configure Claude Code to route LLM calls through OpenRouter, enabling access to models from multiple providers through a single API.

OpenRouter is an API gateway that gives you access to LLMs from multiple providers (Anthropic, OpenAI, DeepSeek, Meta, Google, etc.) through a single endpoint
- Use Claude Code with non-Anthropic models like DeepSeek or OpenAI GPT
- Or mix models from different providers depending on the task
For a general introduction to Claude Code, see the Claude Code setup guide
This guide covers:
- Setting up OpenRouter, with and without BYOK (Bring Your Own Key)
- Testing the API connection
- Configuring Claude Code to route through OpenRouter

What is OpenRouter and BYOK#

OpenRouter acts as a proxy between your client and LLM providers
- You send requests to https://openrouter.ai/api/v1
- OpenRouter routes them to the appropriate provider
BYOK (Bring Your Own Key) lets you use your existing API keys from providers (Anthropic, OpenAI, etc.) instead of buying credits from OpenRouter
- Useful if you already have subscriptions or credits with a provider
- Register your keys with OpenRouter and traffic routes through your own accounts

Step 1: Configure OpenRouter with BYOK#

Generate an OpenRouter API key (sk-or-v1-...) for your client applications
To bring your own keys, create API keys for your providers at:
- Anthropic Console
- OpenAI API Keys
Provider Environment Variable Key Format

Anthropic ANTHROPIC_API_KEY sk-ant-api03-...

OpenAI OPENAI_API_KEY sk-proj-...
Note that Anthropic encodes in the key ...-api-03-... since you need to use metered use
Register your BYOK keys with OpenRouter at: BYOK Settings

Step 2: Test the OpenRouter API#

List available models through OpenRouter:

> curl https://openrouter.ai/api/v1/models \
    -H "Authorization: Bearer $OPENROUTER_API_KEY" \
    | jq '.data[].id' | sort

Test a simple completion:

> curl https://openrouter.ai/api/v1/chat/completions \
    -H "Authorization: Bearer $OPENROUTER_API_KEY" \
    -H "Content-Type: application/json" \
    -d '{
      "model": "openai/gpt-5",
      "messages": [
        {
          "role": "user",
          "content": "Reply with the word OK"
        }
      ]
    }'

Step 3: Configure Claude Code for OpenRouter#

Claude Code uses Anthropic's SDK, which connects to api.anthropic.com by default
To route through OpenRouter, set these environment variables:

Send the API requests from Claude Code to OpenRouter using their API and key

> export ANTHROPIC_BASE_URL=https://openrouter.ai/api
> export ANTHROPIC_AUTH_TOKEN=$OPENROUTER_API_KEY

Select the models to use for the different Claude Code tier (e.g., using OpenAI GPT-5 using OpenRouter)

> export ANTHROPIC_DEFAULT_HAIKU_MODEL=openai/gpt-5
> export ANTHROPIC_DEFAULT_OPUS_MODEL=openai/gpt-5
> export ANTHROPIC_DEFAULT_SONNET_MODEL=openai/gpt-5

Unset the direct Anthropic key to avoid conflicts
```
> unset ANTHROPIC_API_KEY
```

Verify the environment is configured correctly:

> env | grep ANT | sort
ANTHROPIC_AUTH_TOKEN=sk-or-v1-...
ANTHROPIC_BASE_URL=https://openrouter.ai/api
ANTHROPIC_DEFAULT_HAIKU_MODEL=openai/gpt-5
ANTHROPIC_DEFAULT_OPUS_MODEL=openai/gpt-5
ANTHROPIC_DEFAULT_SONNET_MODEL=openai/gpt-5

Testing Connection to OpenRouter with the Anthropic Python SDK#

Run the script .claude/test_openrouter_api.py to verify that OpenRouter responds when using the Anthropic API

> .claude/test_openrouter_api.py
Endpoint : https://openrouter.ai/api
Model    : openai/gpt-5
Testing...

SUCCESS

Step 4: Using Different Models#

Once OpenRouter is configured, point Claude Code's model tiers at whatever models you want
For help choosing which model fits your workflow, see How to Compare and Choose LLM Models:

A set up I personally like based on the analysis above

> export ANTHROPIC_DEFAULT_HAIKU_MODEL=deepseek/deepseek-v4-flash
> export ANTHROPIC_DEFAULT_SONNET_MODEL=anthropic/haiku-4.5
> export ANTHROPIC_DEFAULT_OPUS_MODEL=anthropic/sonnet-4.6

Verify the current configuration:

> env | grep ANTH | sort
ANTHROPIC_AUTH_TOKEN=sk-or-v1-...
ANTHROPIC_BASE_URL=https://openrouter.ai/api
ANTHROPIC_DEFAULT_HAIKU_MODEL=deepseek/deepseek-v4-flash
ANTHROPIC_DEFAULT_OPUS_MODEL=anthropic/haiku-4.5
ANTHROPIC_DEFAULT_SONNET_MODEL=anthropic/haiku-4.5

Test Claude Code#

Verifying the Model in Claude Code#

Once Claude Code launches note that the model is the one mapped on Haiku (in the case above, it's deepseek-v4-flash) and the access is API Usage Billing (since we are not going through a Claude plan)
```
> claude ...
 ▐▛███▜▌   Claude Code v2.1.158
▝▜█████▛▘  deepseek/deepseek-v4-flash · API Usage Billing
  ▘▘ ▝▝    ~/src/xyz
```
Check which model is active with the /model command:

``` ❯ /model

Select model
   Switch between Claude models. Your pick becomes the default for new
   sessions.

       1. Default (recommended)         Use the default model (currently
          anthropic/haiku-4.5) · $5/$25 per Mtok
       2. anthropic/haiku-4.5           Custom Opus model
       3. anthropic/haiku-4.5           Custom Sonnet model
     ❯ 4. deepseek/deepseek-v4-flash ✔  Custom Haiku model
```

Monitoring and Troubleshooting#

Check OpenRouter Logs for request history, latency, and error details
For an alternative workflow using OpenRouter with Simon Willison's LLM CLI, see How to Use OpenRouter with LLM CLI
Refer to the OpenRouter API documentation for the full API specification
Common issues:
- Authentication errors: Make sure ANTHROPIC_AUTH_TOKEN is your OpenRouter key, not your Anthropic key
- Model not found: Check the exact model ID on OpenRouter's models page (e.g., deepseek/deepseek-v4-flash, not deepseek-v4-flash)
- Rate limiting: OpenRouter applies rate limits per provider
- Key conflicts: Unset ANTHROPIC_API_KEY when using OpenRouter since it can conflict with ANTHROPIC_AUTH_TOKEN

Using cc#

This repository also ships a cc wrapper script that automates the environment variable setup for OpenRouter
See The cc Convenience Wrapper)

Provider	Environment Variable	Key Format
Anthropic	`ANTHROPIC_API_KEY`	`sk-ant-api03-...`
OpenAI	`OPENAI_API_KEY`	`sk-proj-...`