# Developer FAQ: API keys, base URL, billing > Answers to common BigModel.io questions about base URLs, API keys, groups, model IDs, balance, free models, rate limits, errors, refunds and data handling. Web version: https://bigmodel.io/docs/faq · Updated 2026-10-18 Short answers to the questions we get most. Each links to the page with details. ## Which base URL do I use? OpenAI-style clients and SDKs: `https://bigmodel.io/v1`. Anthropic-style clients (Claude Code, Anthropic SDK): `https://bigmodel.io` without `/v1`, because the client adds `/v1/messages` itself. Adding `/v1` twice gives a 404. See the [quickstart](https://bigmodel.io/docs/quickstart). ## I lost my API key. Do I need a new one? No. Open [API Keys](https://bigmodel.io/keys) and use the copy button; it copies the full key at any time. If you think the key leaked, disable it and create a new one. See [Authentication](https://bigmodel.io/docs/authentication). ## Why does my model return 503 "No available accounts"? That is how BigModel.io reports a model the key cannot use: either the ID is mistyped, or the key's group does not serve it. Copy the ID from the [catalog](https://bigmodel.io/models) (lowercase, dots kept) and check `GET /v1/models`. See [Errors](https://bigmodel.io/docs/errors). ## What is a group, and why does my key get 404 on /v1/messages? Every key belongs to a group, which decides the models, endpoints and price multiplier. GLM and DeepSeek groups serve only `/v1/chat/completions`, so Claude Code (Messages) and Codex CLI (Responses) need a key from a group that serves those endpoints. ## Can I use Claude Code with GLM, Kimi or DeepSeek? Yes. Set `ANTHROPIC_BASE_URL=https://bigmodel.io`, `ANTHROPIC_AUTH_TOKEN` to your key and `ANTHROPIC_MODEL` to a model ID, using a key whose group serves `/v1/messages`. See [Claude Code](https://bigmodel.io/docs/claude-code). ## Why do I get 403 when my balance is zero? Paid models and endpoints such as `GET /v1/models` need a balance above zero; with a zero balance they return 403. Models priced at $0 stay callable on the chat, messages and responses endpoints, with a per-minute limit while the balance is zero. [Top up](https://bigmodel.io/purchase) any amount to use paid models. See [Trial credit & free models](https://bigmodel.io/docs/free-credits). ## How am I charged? In USD per token: input, output, cache read and cache write tokens at the model's price, times your group's multiplier. Failed requests are not charged. Every charge is listed in [Usage](https://bigmodel.io/usage). See [Billing & top-ups](https://bigmodel.io/docs/billing). ## How can I pay? In [Billing](https://bigmodel.io/purchase). Methods are shown at checkout, for example Alipay and USDT (TRC20, ERC20, BEP20). Your balance is always in USD. ## Can I get a refund? Unused plans can be refunded within 24 hours of purchase. USDT payments are not refundable, and promotional credit has no cash value. See the [refund policy](https://bigmodel.io/legal/refund-policy). ## Can I limit how much one key spends? Yes. Each key can have a USD quota and 5-hour, 1-day and 7-day spend windows, plus an expiry date and IP rules. Windows are fixed, not rolling. See [Limits & rate limits](https://bigmodel.io/docs/rate-limits). ## I am getting 429 errors. What should I do? Read the error code. `API_KEY_QUOTA_EXHAUSTED`, `USAGE_LIMIT_EXCEEDED` and window messages ("限额已用完") will not clear by retrying. "Too many pending requests" and "Concurrency limit exceeded" mean too many parallel requests; RPM limits send `Retry-After`. See [Limits & rate limits](https://bigmodel.io/docs/rate-limits). ## Does my OpenRouter code work? Mostly, by changing the base URL and key. `provider/model` IDs, the `models` fallback array and `HTTP-Referer` / `X-Title` are accepted; the `provider` object, `transforms` and `route` are ignored. See [Migrate from OpenRouter](https://bigmodel.io/docs/migrate-from-openrouter). ## Are my prompts stored or used for training? Prompts and outputs of successful requests pass through to the model provider and are not kept after delivery. Bodies of failed requests are kept up to 30 days, with credentials removed, for troubleshooting. Prompts are not used for training. The provider may retain data under its own terms. See the [privacy policy](https://bigmodel.io/legal/privacy-policy). ## How do I check whether a model is up? The [status page](https://bigmodel.io/status) shows availability, latency and time to first token per model over 24 hours, 7 days and 30 days. ## How do I contact support? Use [Support](https://bigmodel.io/support) in the console or email support@bigmodel.io. Replies within one business day. Include the `X-Request-ID`, time, model, endpoint and the error body; see [Errors](https://bigmodel.io/docs/errors). ## Troubleshooting **My question is not here.** Search the docs, then ask [Support](https://bigmodel.io/support). **An answer here disagrees with what I see.** Trust the error message and console, and tell us so we can fix the page. ## Related - [Overview](https://bigmodel.io/docs/overview) - [API quickstart](https://bigmodel.io/docs/quickstart) - [Errors](https://bigmodel.io/docs/errors)