Integration
Cheaperinference Provider
By OpenClaw provider plugin for Cheaper Inference, an OpenAI-compatible LLM API.
Install
openclaw plugins install clawhub:openclaw-cheaperinference-providerREADME
# Cheaper Inference provider for OpenClaw
This plugin adds [Cheaper Inference](https://cheaperinference.com) as a model provider to OpenClaw.
Each model costs 15–60% less than the list price of its lab.
Cheaper Inference has an OpenAI-compatible API.
One API key gives access to models from many labs.
| Property | Value |
| --------------- | --------------------------------------- |
| Provider id | `cheaperinference` |
| Package | `openclaw-cheaperinference-provider` |
| API key env var | `CHEAPER_INFERENCE_API_KEY` |
| Onboarding flag | `--auth-choice cheaperinference-api-key` |
| Direct CLI flag | `--cheaperinference-api-key <key>` |
| API | OpenAI Chat Completions |
| Base URL | `https://api.cheaperinference.com/v1` |
| Default model | `cheaperinference/gpt-5.4-mini` |
## Requirements
- OpenClaw 2026.9.6 or later.
- A Cheaper Inference API key.
Get a key at <https://cheaperinference.com/signup>.
The key starts with `ci_live_`.
## Install
```bash
openclaw plugins install clawhub:openclaw-cheaperinference-provider
```
A running Gateway loads the plugin automatically.
If the Gateway does not run, the plugin loads at the next start.
## Set the API key
Use one of these methods.
Method 1: run onboarding and select **Cheaper Inference**.
```bash
openclaw onboard --auth-choice cheaperinference-api-key
```
The plugin sends one request to `GET /v1/models` to check the key.
If the API rejects the key, OpenClaw does not save it.
If the API is not available, OpenClaw saves the key and shows a note.
Method 2: sign in for one provider only.
```bash
openclaw models auth login --provider cheaperinference --set-default
```
Method 3: set the environment variable for the Gateway process.
```bash
export CHEAPER_INFERENCE_API_KEY="<your-api-key>"
```
Method 4: non-interactive setup, for example in a script.
```bash
openclaw onboard --non-interactive --accept-risk --skip-health \
--mode local \
--auth-choice cheaperinference-api-key \
--cheaperinference-api-key "$CHEAPER_INFERENCE_API_KEY"
```
## Select a model
Show the models:
```bash
openclaw models list --provider cheaperinference
```
Set the default model:
```bash
openclaw models set cheaperinference/gpt-5.4-mini
```
Write the model id after the `cheaperinference/` prefix.
Examples:
- `cheaperinference/gpt-5.4-mini`
- `cheaperinference/gpt-5.4`
- `cheaperinference/claude-sonnet-5`
- `cheaperinference/gemini-3.1-pro`
- `cheaperinference/deepseek-v4-flash`
- `cheaperinference/glm-5.3`
The full model list is at <https://cheaperinference.com/#models>.
## Config example
`~/.openclaw/openclaw.json`:
```json5
{
env: { vars: { CHEAPER_INFERENCE_API_KEY: "<your-api-key>" } },
agents: {
defaults: {
model: {
primary: "cheaperinference/gpt-5.4-mini",
fallbacks: ["cheaperinference/claude-sonnet-5"],
},
},
},
}
```
## How the model list works
When a key is set, the plugin reads `GET /v1/models`.
It keeps only chat models (rows with `type: "text"`).
It does not show image or video models.
For each model, the plugin reads these fields:
| API field | OpenClaw field |
| ----------------------------------------- | ------------------------ |
| `context_length` | context window |
| `max_output_tokens` | output token limit |
| `capabilities.vision` | image input |
| `capabilities.reasoning` | reasoning |
| `pricing.input_per_million` | input cost (USD / 1M) |
| `pricing.output_per_million` | output cost (USD / 1M) |
| `pricing.cache_read_input_per_million` | cache read cost (USD / 1M) |
| `pricing.cache_write_input_per_million` | cache write cost (USD / 1M) |
The plugin keeps the list in memory for 60 seconds.
If the request fails, the plugin uses a small offline list:
`gpt-5.4-mini`, `gpt-5.4`, `claude-sonnet-5`, `gemini-3.1-pro`.
The offline list has no prices.
A model id that is not in the list also works.
OpenClaw uses a 128,000-token context window and an 8,192-token output limit for it.
## Request format
The plugin sends requests to `/v1/chat/completions`.
It sends the output limit as `max_tokens`.
It sends the system prompt with the `system` role.
It does not send the `store` field.
For Claude models (ids that start with `claude-`), it adds Anthropic `cache_control` markers.
With these markers, the API caches the conversation, not only the system prompt.
## Data and network
The plugin sends your API key only to `https://api.cheaperinference.com`.
If you set a different `baseUrl` for this provider, the plugin does not read the model list.
The plugin does not read environment variables directly.
OpenClaw gives the key to the plugin.
## Limits
- Chat models only.
The plugin does not add embeddings, speech, image generation, or video generation.
## Links
- Documentation: <https://cheaperinference.com/docs>
- Models: <https://cheaperinference.com/#models>
- Sign up: <https://cheaperinference.com/signup>
## License
MIT
integration
Comments
Sign in to leave a comment