# Use BigModel.io with LobeChat

> Connect LobeChat (LobeHub) to BigModel.io through the settings UI or self-hosting env vars OPENAI_PROXY_URL=https://bigmodel.io/v1 and OPENAI_MODEL_LIST, with fixes for common errors.

Web version: https://bigmodel.io/docs/lobechat · Updated 2026-10-18

LobeChat (now LobeHub) can use BigModel.io as its OpenAI provider, as a custom provider, or through environment variables on a self-hosted instance. All three use the OpenAI request format on `https://bigmodel.io/v1`.

| Setting | Value |
|---|---|
| API key | your BigModel.io API key (`sk-...`) |
| Proxy URL (OpenAI format) | `https://bigmodel.io/v1` |
| Proxy URL (Anthropic format) | `https://bigmodel.io` (no `/v1`) |
| Models | model IDs from the [model catalog](https://bigmodel.io/models) |

## Option 1: settings UI

1. Open **Settings → AI Providers**.
2. Either open **OpenAI**, or choose **Create custom AI provider** and enter a Provider ID such as `bigmodel` with request format **openai**.
3. Enter your key from [API Keys](https://bigmodel.io/keys) and set **Proxy URL** to `https://bigmodel.io/v1`.
4. Click **Check** to test the connection.
5. Add or enable the models you want, using exact IDs such as `glm-5.3`, `kimi-k3` or `deepseek-v4-pro`.

A custom provider keeps OpenAI's own setup free. Request format **anthropic** with Proxy URL `https://bigmodel.io` also works, but only with keys whose group serves `/v1/messages`.

## Option 2: self-hosting environment variables

For a self-hosted instance, set these in your `.env` or container environment:

```bash
OPENAI_API_KEY=sk-your-bigmodel-key
OPENAI_PROXY_URL=https://bigmodel.io/v1
OPENAI_MODEL_LIST="-all,+glm-5.3=GLM-5.3<128000:fc>,+kimi-k3=Kimi K3<128000:fc>,+deepseek-v4-pro=DeepSeek V4 Pro<128000:fc>"
```

`OPENAI_MODEL_LIST` syntax:

- `-all` hides the built-in OpenAI models, which BigModel.io does not serve.
- `+id=Display Name` adds a model; `id` must be the exact catalog ID.
- `<maxToken:vision:reasoning:search:fc:file>` declares the context size and abilities. Use `fc` only for models with tool calling and `vision` only for image input. The `128000` values above are placeholders; take the context window from each model page.

For the Anthropic format instead, set `ANTHROPIC_API_KEY` and `ANTHROPIC_PROXY_URL=https://bigmodel.io`. Restart the container after changing env vars.

## Verify the setup

Start a new chat, pick one of the BigModel.io models in the model selector and send "ping". You should get a reply, and the request should appear in [Usage](https://bigmodel.io/usage).

## Troubleshooting

**Check fails with 403 `INSUFFICIENT_BALANCE`.** Your balance is $0 or below. Top up in [Billing](https://bigmodel.io/purchase).

**Check fails because the default model is unavailable.** LobeChat may check with a built-in OpenAI model name that your group does not serve (503 `api_error`: `No available accounts ...`). Add your own model, select it for the check, and try again.

**404 on every request.** The Proxy URL is wrong. OpenAI format needs `https://bigmodel.io/v1`; Anthropic format needs `https://bigmodel.io`.

**404 `not_found_error`: `... /v1/messages is not available`.** You chose the Anthropic format with a key from a Chat Completions-only group (GLM or DeepSeek). Use the openai format.

**Built-in GPT models still appear.** `OPENAI_MODEL_LIST` does not start with `-all`, or the container was not restarted.

**401 `INVALID_API_KEY`.** The key has spaces, was deleted, or `OPENAI_API_KEY` still holds an OpenAI key.

## Related

- [Cherry Studio](https://bigmodel.io/docs/cherry-studio) for a desktop chat client.
- [Dify](https://bigmodel.io/docs/dify) for building apps and workflows.
- [Tool calling](https://bigmodel.io/docs/tool-calling) and [Image input](https://bigmodel.io/docs/vision) for which abilities to declare.