> ## Documentation Index > Fetch the complete documentation index at: https://docs.laozhang.ai/llms.txt > Use this file to discover all available pages before exploring further. # Text generation API: protocols and models > Generate text with LaoZhang API: pick the OpenAI-compatible, Anthropic Messages, or Gemini format, compare models, and read streamed replies. LaoZhang API serves text models in three request formats with the same API key: OpenAI-compatible Chat Completions, Anthropic Messages, and Gemini's native generateContent. For a new project, use Chat Completions; it works with the largest number of text models. | Item | Details | | - | - | | API host | `https://api.laozhang.ai` | | New project | Send `model` and `messages` to `POST /v1/chat/completions` and read the reply from `choices[0].message.content` | | Existing SDK app | If it already uses the Anthropic SDK, Claude Code, or the Google Gen AI SDK, keep its request format and change only the base URL and API key | | Formats per model | The **Endpoints** column in the [model catalog](/en/models) | | Prices | See [Common text models](#common-text-models) | ## Choose a protocol * **New project, or a client built on the OpenAI SDK or format**: use OpenAI-compatible Chat Completions. See the [OpenAI protocol reference](/en/api-reference/chat-completions). * **Anthropic SDK, Claude Code, or another Messages client**: use Anthropic Messages. See the [Claude protocol reference](/en/api-reference/claude). * **Google Gen AI SDK, or code that relies on Gemini's native fields**: use the Gemini format. See the [Gemini protocol reference](/en/api-reference/gemini). Each format has its own endpoint: ```text theme={null} OpenAI-compatible POST https://api.laozhang.ai/v1/chat/completions Anthropic Messages POST https://api.laozhang.ai/v1/messages Gemini native POST https://api.laozhang.ai/v1beta/models/{model}:generateContent ``` When you use an SDK, set the OpenAI SDK's base URL to `https://api.laozhang.ai/v1`. The Anthropic SDK and the Google Gen AI SDK take the host root, `https://api.laozhang.ai`, and add the path themselves. Claude models have not been listed on LaoZhang API since September 24, 2026. We will post an [announcement](/en/changelog) if they return. The Anthropic Messages endpoint still serves the models marked for it in the pricing configuration, including: * `glm-5.2` (token groups `default` and `claude_code`) * `deepseek-v4-pro` * `deepseek-v4-flash` The request and response structures differ, so switching formats means changing both the request code and the parser: | | OpenAI-compatible | Anthropic Messages | Gemini native | | - | - | - | - | | Model ID | `model` in the body | `model` in the body | URL path | | Input | `messages` | `messages` and required `max_tokens` | `contents[].parts[]` | | System prompt | `system` role message | Top-level `system` | `systemInstruction` | | Reply text | `choices[0].message.content` | `text` blocks in `content[]` | `text` in `parts[]` | | Streaming | `"stream": true` | `"stream": true` | `streamGenerateContent` | A model marked with two formats, such as `gemini-3.8-flash` (Chat Completions and Gemini native), accepts either one, but each request must use a single format's fields. ## Common text models Prices below come from LaoZhang API's public pricing as of September 24, 2026. They are list prices for the `default` token group, in US dollars per 1M tokens. The [model catalog](/en/models) has cache-read prices, other models, and other token groups. | Model ID | Formats | Input / output | | - | - | - | | `gpt-6-luna` | OpenAI-compatible | \$0.1 / \$0.5 | | `gpt-6-sol` | OpenAI-compatible | \$2 / \$10 | | `gpt-6.1-sol` | OpenAI-compatible | \$2 / \$10 | | `gpt-6-astra` | OpenAI-compatible | \$10 / \$50 | | `gpt-5.6-luna` | OpenAI-compatible | \$0.2 / \$1.2 | | `gpt-5.6-terra` | OpenAI-compatible | \$2 / \$12 | | `gpt-5.6-sol` | OpenAI-compatible | \$4 / \$20 | | `gpt-5.4-mini` | OpenAI-compatible | \$0.75 / \$4.5 | | `gpt-5.4` | OpenAI-compatible | \$2.5 / \$15 | | `gemini-4-argon` | OpenAI-compatible, Gemini native | \$2 / \$10 | | `gemini-3.8-flash` | OpenAI-compatible, Gemini native | \$0.75 / \$3.75 | | `gemini-3.6-flash` | OpenAI-compatible, Gemini native | \$1.5 / \$7.5 | | `gemini-3.5-flash-lite` | OpenAI-compatible, Gemini native | \$0.3 / \$2.502 | | `glm-5.2` | OpenAI-compatible, Anthropic Messages | See catalog | | `deepseek-v4-pro` | OpenAI-compatible, Anthropic Messages | See catalog | | `deepseek-v4-flash` | OpenAI-compatible, Anthropic Messages | See catalog | | `o3-pro` | Responses | See catalog | * GPT-6 models, GPT-5.6 models, and `gpt-5.4` charge a higher tier for long inputs. The catalog's [tiered pricing](/en/models#tiered-pricing) table lists the thresholds and prices. * To choose among the three GPT-6 models, and for OpenAI's rules on their parameters and tool calling in Chat Completions, see the [GPT-6 launch announcement](/en/announcements/gpt-6-launch-2026-09). * The OpenAI models in the table also work with the Responses endpoint, `POST /v1/responses`, and the `o3-pro` series supports only Responses. * Tool calling with GPT-6 Astra or GPT-6.1 Sol requires Responses. For GPT-6.1 Sol's other parameter differences, see the [GPT-6.1 Sol launch announcement](/en/announcements/gpt-6-1-sol-2026-10). See [OpenAI models on LaoZhang API](/en/api-reference/openai) for examples. * For Gemini 4 Argon's thinking levels and launch pricing, see the [Gemini 4 Argon launch announcement](/en/announcements/gemini-4-argon-2026-10). * To send images to a vision model, see the [Image understanding API](/en/api-capabilities/vision-understanding). ## Send your first request Create an API key in [Token management](https://api.laozhang.ai/token). Set its billing mode to Usage first (按量优先, recommended) or Usage-based (按量计费), and make sure its token group can call the model you want. Then export the key: ```bash theme={null} export LAOZHANG_API_KEY="YOUR_LAOZHANG_API_KEY" ``` These examples send one message to `gpt-5.4-mini` in the OpenAI-compatible format. Install the SDK before you run them: * Python: run `python -m pip install openai`. * Node.js: run `npm install openai`, save the code as `chat.mjs`, and run it with `node chat.mjs`. ```bash theme={null} curl --fail-with-body --max-time 120 \ "https://api.laozhang.ai/v1/chat/completions" \ -H "Authorization: Bearer $LAOZHANG_API_KEY" \ -H "Content-Type: application/json" \ -d '{ "model": "gpt-5.4-mini", "messages": [{"role": "user", "content": "Reply only with: connected"}] }' ``` ```python theme={null} import os from openai import OpenAI client = OpenAI( api_key=os.environ["LAOZHANG_API_KEY"], base_url="https://api.laozhang.ai/v1", timeout=120.0, max_retries=0, ) response = client.chat.completions.create( model="gpt-5.4-mini", messages=[{"role": "user", "content": "Reply only with: connected"}], ) print(response.choices[0].message.content) ``` ```javascript theme={null} import OpenAI from "openai"; const client = new OpenAI({ apiKey: process.env.LAOZHANG_API_KEY, baseURL: "https://api.laozhang.ai/v1", timeout: 120_000, maxRetries: 0, }); const response = await client.chat.completions.create({ model: "gpt-5.4-mini", messages: [{ role: "user", content: "Reply only with: connected" }], }); console.log(response.choices[0].message.content); ``` The reply should be `connected`. To try another model, change `model` to an ID whose **Endpoints** include OpenAI-compatible. The SDK examples turn off automatic retries so that your application decides when to send a request again. ## Stream the reply Add `"stream": true` to receive the reply as server-sent events while the model writes it. Each format streams a little differently: * **OpenAI-compatible**: append each `choices[0].delta.content` in order and skip chunks that have no `choices` or `content`. The stream ends with `data: [DONE]`. [Stream a response](/en/api-reference/chat-completions#stream-a-response) has complete cURL and Python examples and the raw events. * **Anthropic Messages**: text arrives as `delta.text` in `content_block_delta` events, and the reply is complete only after `message_stop`. See [Stream the response](/en/api-reference/claude#stream-the-response). * **Gemini native**: replace `generateContent` with `streamGenerateContent` and add `?alt=sse`. Each event carries a partial response with `candidates`; see [Stream the reply](/en/api-reference/gemini#stream-the-reply). * If the connection closes before the end marker, the text you received is incomplete; treat the request as failed or send it again. With cURL, add `-N` to turn off output buffering. ## Troubleshoot common errors * **404 with `/v1/v1` in the URL**: the SDK base URL repeats part of the path. Use `https://api.laozhang.ai/v1` for the OpenAI SDK and `https://api.laozhang.ai` for the Anthropic and Google Gen AI SDKs. * **Model not available, 403, or 404**: copy the exact model ID from the catalog, confirm that your key's token group includes the model, and check that the model supports the format you are sending. For example, `o3-pro` works only through Responses, tool calling with `gpt-6-astra` requires Responses, and GPT models do not accept Gemini native requests. See [model availability](/en/faq/model-availability). * **400**: a Messages request without `max_tokens` is a common cause. Otherwise, go back to the minimal request, then add optional fields one at a time and fix the field the error names. * **401**: the API key is wrong, the header is malformed, or the environment variable is not set in the current shell. * **Empty reply**: * Chat Completions: check `message.tool_calls` and `finish_reason`. * Messages: read `content[]` by block `type` instead of taking `content[0]`. * Gemini: check `finishReason` and `promptFeedback`. * **Truncated reply**: any of these values means the reply hit the output limit. Raise the model's output limit or shorten the input. * Chat Completions: `finish_reason` of `length`. * Messages: `stop_reason` of `max_tokens`. * Gemini: `finishReason` of `MAX_TOKENS`. * **429**: read the error body to tell rate or concurrency limits apart from insufficient balance. If you still have balance, see [why a request can fail with remaining balance](/en/faq/balance-insufficient). ## Before you go to production 1. Pin the model ID and protocol in configuration, and test a model change like any other release. 2. Run a sample of your own prompts through two or three candidate models and compare answer quality, latency, and token usage. [Choose a model](/en/api-reference/chat-completions#choose-a-model) describes the comparison. 3. Test normal requests, streaming, and error handling. Keep retries in one layer, either the SDK or your application. 4. In [call logs](https://api.laozhang.ai/log), check each request's model, input and output tokens, and charge. ## Related documentation * [OpenAI protocol: Chat Completions](/en/api-reference/chat-completions) * [Use OpenAI models with LaoZhang API](/en/api-reference/openai) * [Claude protocol: Anthropic Messages](/en/api-reference/claude) * [Gemini protocol: generateContent](/en/api-reference/gemini) * [API integration guide](/en/api-manual): keys, base URLs, Responses, and images * [Model catalog and pricing](/en/models) * Upstream documentation: [OpenAI text generation](https://developers.openai.com/api/docs/guides/text), [Anthropic Messages API](https://platform.claude.com/docs/en/api/messages), and [Gemini text generation](https://ai.google.dev/gemini-api/docs/text-generation) This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.